Compare commits

..
Author SHA1 Message Date
jcwalker3andClaude Opus 4.8 0104a76eea docs(remote-mcp): restamp the commit the #956 anchors resolve at
The execution-authorization gate shifted two cited lines in
mcp_daemon_guard.py, so the fixture and the document must name the commit
where the re-anchored citations actually resolve. Both still named
c1626081, where the two moved anchors no longer point at the claimed code.

Point both at a143cd06. This commit changes no .py file, so every anchor
that resolves at a143cd06 resolves here too. The historical note about
#930's drift between 7bf4f125 and aad5c8b4 stays as written.

Refs #931, #956. Addresses review 635.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
2026-07-28 11:17:47 -05:00
jcwalker3andClaude Opus 4.8 a143cd065b fix(transport): gate transport execution behind commissioning (review 635)
Addresses B1 and B2 from formal review 635 on PR #966. Both are one defect:
the seam conflated two questions that must stay separate.

  recognition:  "is this an identifier this system knows, and may it be
                 bound, pinned and recorded?"
  execution:    "is this entrypoint commissioned to actually serve on it?"

SUPPORTED_TRANSPORTS answered the first and was then used to answer the
second, so registering streamable-http was the same act as executing it.
With mcp.run fed the bound value, GITEA_MCP_TRANSPORT=streamable-http
reached FastMCP.run -> run_streamable_http_async and started a uvicorn
HTTP server over the whole tool surface with auth=None, no TLS and no
per-request principal. That endpoint is owned by #938 and is an explicit
non-goal of #931. B2 is why B1 was reachable: bound_transport was read
only by reporting sites, so no decision anywhere depended on it.

The fix adds an execution-authorization layer rather than removing the
identifier or mapping it to stdio, both of which would have destroyed the
seam #931 exists to create.

mcp_transport_config gains EXECUTABLE_TRANSPORTS, a strict subset of
SUPPORTED_TRANSPORTS containing only the local transport, plus
TRANSPORT_EXECUTION_OWNER recording that #938 commissions the remote
listener. assess_transport_execution returns a structured verdict:
transport, recognized, executable, allowed, blocker_kind, owner_issue,
reasons and exact_next_action.

mcp_daemon_guard gains assess_serve_authorization, which is the decision
that consumes bound_transport and is what stops it being reporting-only
metadata, and authorize_transport_execution, which enforces two ordered
boundaries: assert_transport_bound first, so an unbound runtime keeps its
pre-existing #695 failure and reason code; then execution authorization,
so a registered but uncommissioned transport is refused before any
listener exists and before any tool can dispatch. TransportExecutionError
subclasses UnsanctionedRuntimeError, so every existing fail-closed handler
still catches it while a caller that cares can distinguish the two cases.
native_runtime_status now reports executable_transports, serve_authorized
and the full serve_authorization verdict.

The entrypoint serves through authorize_transport_execution.

Net effect. stdio: unchanged, still binds and still reaches the runner.
streamable-http: still recognized, still validated, still pinned, still
recorded in decision-lock and audit provenance -- and refused at the serve
boundary with blocker transport_listener_not_commissioned naming #938.
Unregistered identifiers still fail earlier, at bind validation. Unbound
execution keeps the #695 contract. Rebind idempotence and conflict
rejection, session isolation, provenance and secret handling are untouched.

Tests: 63 in tests/test_issue_931_transport_bind_seam.py, up from 42. New
coverage proves stdio reaches the real runner; the remote identifier binds
and is recorded yet never reaches run(); no socket is bound during the
refusal, asserted with a socket.bind tripwire; the refusal names the
transport, the unmet requirement and #938; the decision genuinely consumes
bound_transport, proven by flipping only that value; no mutation is
authorized after denial; verdicts do not leak across runtimes; and the
refusal carries no credential material. Per review 635 the two file-list
scans are now globs over the production surface, and a new test asserts
there is exactly one mcp.run site and that it routes through the guard.

#956 anchors re-anchored for the two lines this change shifted,
mcp_daemon_guard.py 178->195 and 519->583. No other anchor moved.

Full suite from the branches/ worktree: 28 failed, 5864 passed, 6 skipped,
1047 subtests. Base 9b80e75c: 28 failed, 5801 passed, 6 skipped, 1047
subtests. Failing test IDs are identical sets in both directions; the +63
passed are this module.

No listener, socket, authentication, TLS or principal code is added. #938,
#957, #958 and #959 remain unimplemented.

Refs #931, #938. Addresses review 635.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
2026-07-28 11:15:00 -05:00
jcwalker3andClaude Opus 4.8 2fb835a1aa docs(remote-mcp): stamp the commit the #956 anchors were re-derived at
The anchor fixture records the commit its anchors were taken at, and the
threat model must state the same commit, so a reviewer can resolve every
cited line at a named revision. Both still named aad5c8b4, where the
anchors no longer resolve after the transport seam shifted the lines they
point at.

Point both at c1626081, the seam commit whose tree the anchors were
re-derived from. This commit changes no .py file, so every anchor that
resolves at c1626081 resolves here too. The historical note about #930's
drift between 7bf4f125 and aad5c8b4 stays as written -- it is the reason
the guard exists.

Refs #931, #956.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
2026-07-28 10:01:01 -05:00
jcwalker3andClaude Opus 4.8 c162608175 feat(transport): add transport-neutral MCP bind seam
Closes #931.

The transport was a literal passed once at the bottom of the module,
bind_native_mcp_transport(transport="stdio"), and every guard that asks
"is this a trusted native session" resolves that question through the
value bound there. The string was a constant in the authorization chain
rather than configuration, so a remote transport had no way to be
expressed and no way to be told apart from an untrusted offline import.

mcp_transport_config becomes the single source of truth: it owns the
permitted set, the default, and the resolution of deployment
configuration (GITEA_MCP_TRANSPORT) into a validated identifier. Unset
configuration still yields stdio, so existing deployments are unchanged.
streamable-http is accepted so the bind is pluggable; standing up its
listener remains #938. sse is deliberately not registered.

bind_native_mcp_transport now resolves from configuration when no
transport argument is given, validates against that one permitted set
before writing the runtime record, and pins the result. bound_transport
is the shared accessor every transport-aware guard reads; it reports the
pinned value and never the environment, so a post-bind GITEA_MCP_TRANSPORT
change cannot move what a guard observes -- the rule already applied to
the session-state root under #695 AC2. Rebinding the same identifier is
idempotent; rebinding a different one is refused, so two guards can never
disagree within one process. assert_transport_bound fails closed before
mcp.run, so an absent or invalid bind stops the server instead of serving
tools over a transport no guard can name.

The bound identifier is now recorded in durable provenance:
mutation_provenance_fields gains bound_transport, which reaches the
decision lock through mcp_session_state.save_state and the audit records
that already spread those fields. The pre-existing transport field keeps
its trust-class values, so no durable record changes shape.
assess_transport_for_auth_mint reports the bound transport through the
same accessor; its verdict is unchanged.

#956's anchor fixture and threat model are re-anchored for the lines this
change shifts, and the two boundary claims that #931 makes false -- the
"literal stdio bind" in B1 and the stdio contract at mcp_server.py:4 --
are restated. Without this the #956 guard fails, which is what it is for.

Non-goals, untouched: the remote listener (#938), per-request principal
resolution (#932), client-managed provenance policy (#934), TLS and
remote client authentication, role capability sets, repository-binding
semantics.

Tests: tests/test_issue_931_transport_bind_seam.py, 42 tests, covering
default bind, explicit stdio, the accepted non-stdio identifier, an
unregistered identifier, no bind at all, one-value agreement across
guards, post-bind environment tampering, the durable decision-lock
record, the rebind contract, and the #695 protections that must not
regress.

Full suite from the branches/ worktree: 28 failed, 5843 passed, 6
skipped, 1047 subtests. The pinned base 9b80e75c reports 28 failed, 5801
passed, 6 skipped, 1047 subtests. The failing test IDs are identical sets
-- no failure added, none fixed; the +42 passed are this PR's new module.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
2026-07-28 09:59:01 -05:00
sysadmin 9b80e75ca3 Merge pull request 'docs(remote-mcp): threat model, trust boundaries, and decomposition ruling (Closes #956)' (#965) from docs/issue-956-remote-mcp-threat-model into master 2026-07-28 08:19:45 -05:00
sysadminandClaude Opus 4.8 b3de9c941c docs(remote-mcp): threat model, trust boundaries, and decomposition ruling (Closes #956)
#930 inventoried stdio coupling; nothing stated what the adversary is, what
each boundary protects, or why one process may hold credentials for several
services. This adds that document as child 2 of epic #929.

Adds docs/remote-mcp/threat-model.md covering 10 assets, the 5 adversaries
#956 names, 9 trust boundaries, the data flows between them, a 14-entry
per-boundary credential inventory, and an explicit decomposition ruling.

Findings established from live native evidence at this commit:

- Four prgs roles (reviewer, merger, reconciler, controller) resolve to one
  Gitea account, so separation of duty between approving and landing is
  enforced only by which process a call reaches. The mdcps tenant has no role
  separation at all: author, reviewer, and merger share one account.
- Any one role process can resolve every other role's credential.
  gitea_list_profiles reports "credentials present" for other roles because it
  calls resolve_token on each one.
- The Gitea server reads Jenkins and GlitchTip secrets out of the keychain to
  produce the "authenticated" word in gitea_audit_config's service summaries.
- Jenkins and GlitchTip were already decomposed into separate MCP servers; the
  credential references were left behind in the Gitea configuration.

Ruling D1 forbids a single integration process from holding credentials for
unrelated services, with one time-boxed dual-run exception for the local fleet
that expires with #939. D2 requires separation of duty to be credential-backed,
D3 scopes credential resolution to the request principal, and D4 gives
coordination state its own authority. Every #929 child from 2 through 10 is
mapped to the boundary it implements.

Anchors are enforced rather than asserted. #930's inventory anchors into
gitea_mcp_server.py had already drifted between 7bf4f125 and aad5c8b4 with
nothing detecting it, so this change ships the guard that was missing:
docs/remote-mcp/threat-model-anchors.json declares all 58 anchors with the
substring each must contain, and tests/test_issue_956_threat_model.py fails if
any anchor does not resolve, if the document cites an anchor the fixture does
not cover, or if the structural obligations regress.

Documentation only. No server behavior changes.

Tests:
- tests/test_issue_956_threat_model.py: 17 passed.
- Four sabotage probes confirm the validator is not passing vacuously
  (shifted anchor, undeclared citation, broken count tally, and a blanked
  boundary owner). The last two probes exposed real weaknesses in the checks
  themselves, which were fixed: the section slice now stops at the next
  heading, and boundary ownership is read only from mapping table rows.
- Docs-sensitive sweep (17 modules referencing docs/): 559 passed.
- Full suite: 30 failed, 5799 passed, 6 skipped. All 30 reproduce on a clean
  base worktree at aad5c8b4; branch failures are a strict subset of base
  failures. No production file is modified by this change.

Closes #956

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
2026-07-28 04:41:26 -04:00
sysadmin aad5c8b423 Merge pull request 'fix(author): unify the bootstrap and lock_issue issue-lock contract (Closes #953)' (#954) from fix/issue-953-bootstrap-lock-provenance into master
Merges PR #954 at approved head b4c9f55890.

Approval: review 633 APPROVE at b4c9f55890.
Base: master at 82d71b7702.

Closes #953
2026-07-28 02:21:11 -05:00
10 changed files with 2236 additions and 19 deletions
+80
View File
@@ -0,0 +1,80 @@
{
"_comment": [
"Machine-checkable anchor table for docs/remote-mcp/threat-model.md (#956).",
"Every file:line anchor cited in the threat model must appear here, and the",
"source line at that anchor must contain the 'expect' substring.",
"tests/test_issue_956_threat_model.py enforces both directions, so a refactor",
"that shifts a line number fails the suite instead of silently rotting the",
"document. #930's inventory had no such guard and its gitea_mcp_server.py",
"anchors drifted between 7bf4f125 and aad5c8b4."
],
"generated_against_commit": "a143cd065ba06e1a2bdc5143a19ec156e53650ef",
"anchors": [
{"anchor": "gitea_mcp_server.py:24728", "expect": "mcp_daemon_guard.bind_native_mcp_transport()"},
{"anchor": "mcp_daemon_guard.py:49", "expect": "_PRODUCTION_TRANSPORTS = mcp_transport_config.SUPPORTED_TRANSPORTS"},
{"anchor": "mcp_daemon_guard.py:195", "expect": "def bind_native_mcp_transport"},
{"anchor": "irrecoverable_provenance.py:497", "expect": "def assess_transport_for_auth_mint"},
{"anchor": "gitea_mcp_server.py:9132", "expect": "assess_transport_for_auth_mint()"},
{"anchor": "gitea_mcp_server.py:9381", "expect": "assess_transport_for_auth_mint()"},
{"anchor": "mcp_server.py:4", "expect": "The transport is selected by deployment configuration"},
{"anchor": "gitea_mcp_server.py:15415", "expect": "def _is_client_managed_process"},
{"anchor": "gitea_mcp_server.py:15445", "expect": "def _provenance_mutation_block"},
{"anchor": "gitea_mcp_server.py:15453", "expect": "unsupported_manual_launch"},
{"anchor": "gitea_mcp_server.py:19004", "expect": "server_provenance"},
{"anchor": "gitea_mcp_server.py:21445", "expect": "def _check_mcp_runtimes_diagnostics"},
{"anchor": "gitea_mcp_server.py:21465", "expect": "\"ps\", \"-o\", \"pid,lstart,command\""},
{"anchor": "gitea_mcp_server.py:21509", "expect": "\"ps\", \"eww\""},
{"anchor": "gitea_config.py:1172", "expect": "RECOGNIZED_GITEA_ENV_KEYS"},
{"anchor": "gitea_config.py:1233", "expect": "GITEA_CLIENT_MANAGED"},
{"anchor": "gitea_config.py:54", "expect": "ENV_PROFILE = \"GITEA_MCP_PROFILE\""},
{"anchor": "gitea_config.py:97", "expect": "_REVIEW_MERGE_OPS"},
{"anchor": "gitea_config.py:499", "expect": "repository authorization scope"},
{"anchor": "gitea_config.py:956", "expect": "def _keychain_token"},
{"anchor": "gitea_config.py:974", "expect": "def resolve_token"},
{"anchor": "gitea_config.py:1015", "expect": "def keychain_auth"},
{"anchor": "gitea_config.py:294", "expect": "def _validate_identity_auth"},
{"anchor": "mcp_daemon_guard.py:583", "expect": "def assert_keychain_access_allowed"},
{"anchor": "gitea_mcp_server.py:19261", "expect": "def gitea_list_profiles"},
{"anchor": "gitea_mcp_server.py:19312", "expect": "gitea_config.resolve_token(p)"},
{"anchor": "gitea_mcp_server.py:19555", "expect": "def gitea_audit_config"},
{"anchor": "gitea_mcp_server.py:19577", "expect": "service_summaries(config)"},
{"anchor": "gitea_config.py:704", "expect": "def resolve_service"},
{"anchor": "gitea_config.py:837", "expect": "def service_summaries"},
{"anchor": "gitea_config.py:851", "expect": "_keychain_token(auth.get(\"id\"))"},
{"anchor": "gitea_mcp_server.py:17710", "expect": "\"jenkins-mcp\""},
{"anchor": "gitea_mcp_server.py:17716", "expect": "external-mcp"},
{"anchor": "gitea_mcp_server.py:17737", "expect": "\"glitchtip-mcp\""},
{"anchor": "gitea_mcp_server.py:17742", "expect": "external-mcp"},
{"anchor": "mcp_discoverability.py:9", "expect": "EXPECTED_JENKINS_TOOLS"},
{"anchor": "mcp_discoverability.py:17", "expect": "EXPECTED_GLITCHTIP_TOOLS"},
{"anchor": "sentry_incident_bridge.py:36", "expect": "SENTRY_AUTH_TOKEN"},
{"anchor": "sentry_incident_bridge.py:190", "expect": "def resolve_token"},
{"anchor": "sentry_incident_bridge.py:289", "expect": "Authorization"},
{"anchor": "sentry_observability.py:55", "expect": "SENTRY_DSN"},
{"anchor": "master_parity_gate.py:168", "expect": "def capture_startup_parity"},
{"anchor": "master_parity_gate.py:255", "expect": "mutation_safe"},
{"anchor": "gitea_mcp_server.py:19105", "expect": "def gitea_assess_master_parity"},
{"anchor": "gitea_mcp_server.py:193", "expect": "ACTIVE_WORKTREE_ENV"},
{"anchor": "gitea_mcp_server.py:194", "expect": "AUTHOR_WORKTREE_ENV"},
{"anchor": "gitea_mcp_server.py:2351", "expect": "/tmp/gitea_issue_lock.json"},
{"anchor": "gitea_mcp_server.py:10897", "expect": "def gitea_bootstrap_author_issue_worktree"},
{"anchor": "mcp_server.py:13", "expect": "/tmp/mcp_server_stderr.log"},
{"anchor": "issue_lock_store.py:26", "expect": "DEFAULT_LOCK_DIR"},
{"anchor": "issue_lock_store.py:83", "expect": "def session_pointer_path"},
{"anchor": "issue_lock_store.py:98", "expect": "def is_process_alive"},
{"anchor": "mcp_session_state.py:27", "expect": "DEFAULT_STATE_DIR"},
{"anchor": "control_plane_db.py:47", "expect": "DEFAULT_DB_PATH"},
{"anchor": "control_plane_db.py:380", "expect": "mode=0o700"},
{"anchor": "control_plane_db.py:386", "expect": "sqlite3.connect"},
{"anchor": "control_plane_db.py:1145", "expect": "os.getpid()"},
{"anchor": "gitea_mcp_server.py:12804", "expect": "owner_pid_alive"}
]
}
+407
View File
@@ -0,0 +1,407 @@
# Remote-MCP threat model, trust boundaries, and service decomposition
What the adversary is, what each boundary protects, and which services may share a process.
- **Issue:** #956 (Remote-MCP threat model), child of epic #929, cross-linked to #955.
- **Depends on:** #930 (closed) — `docs/remote-mcp/coupling-inventory.md`.
- **Blocks:** #932, #933, #934, #938.
- **Generated against commit:** `a143cd065ba06e1a2bdc5143a19ec156e53650ef` (#931's transport
bind seam). Originally generated against `aad5c8b42361d380a8eeb07b94b90815e594c2c5`
(`master`) and re-anchored when #931 shifted the cited lines.
- **Scope:** documentation only. This child changes no server behavior. It adds one
document, one anchor fixture, and the test that enforces them.
## Relationship to #930
#930 asked *what breaks when the process stops being local*. This document asks *what an
attacker gets, and where we stop them*. The two are deliberately different axes: #930
classifies each coupling as portable, seam, replacement, or cannot-be-remote; this document
classifies each **credential** by blast radius and each **boundary** by what crossing it
requires. An entry can be perfectly portable and still be a trust disaster —
`gitea_config.py:851` is portable Python that reads a CI secret from inside the Gitea server.
### Anchors are enforced, not asserted
Every `file:line` in this document is declared in `docs/remote-mcp/threat-model-anchors.json`
with the substring that must appear at that line, and
`tests/test_issue_956_threat_model.py` fails if any anchor does not resolve or if the
document cites an anchor the fixture does not cover.
This guard exists because #930 did not have one. Its inventory was generated at
`7bf4f125`; by `aad5c8b4` its `gitea_mcp_server.py` anchors had drifted — the transport
bind it cited at line 23750 now lives at `gitea_mcp_server.py:24728`, and its
client-managed provenance anchor at 14588 now lands in an unrelated function. Nothing
failed, because nothing checked. Anchors into a ~24,700-line module rot silently, and a
security document that cannot prove its own citations is worse than none, because it is
trusted.
---
## 1. Assets
What an adversary wants. Ordered by consequence, not by likelihood.
| ID | Asset | Why it matters |
| -- | ----- | -------------- |
| A1 | Merge authority on `Scaled-Tech-Consulting/Gitea-Tools` | This repository *is* the control plane. Code merged here becomes the gate that authorizes every future mutation, so merge authority is self-amplifying: one merge can disable every other control in this document. |
| A2 | Write authority on the `mdcps` tenant | A second, unrelated organization reachable from the same configuration. Compromise here is a cross-organization incident, not an internal one. |
| A3 | The eight Gitea role credentials | Long-lived bearer tokens. Possession is authority; there is no second factor at the API. |
| A4 | Jenkins read access (`mdcps`, enabled) | Build logs routinely carry deployment topology, internal hostnames, and accidentally-echoed secrets. |
| A5 | Error-tracking read access (GlitchTip / Sentry) | Event payloads carry stack frames, request context, and production user data. |
| A6 | Coordination-state integrity | The locks, leases, and review-decision records that make "exactly one owner" true. Corrupting them needs no Gitea credential and produces duplicate or lost work. |
| A7 | The operator's checkout and worktrees | Unmerged code, branch state, and the filesystem the author tools write to. |
| A8 | The macOS login keychain | The meta-credential. Everything in A3, A4, and A5 resolves from it. |
| A9 | Separation of duty between review and merge | The property that no single actor both approves and lands a change. An *asset*, not a control, because it is what the controls exist to produce. |
| A10 | Audit and provenance records | Determine whether an incident is reconstructable. An attacker who can forge provenance makes an intrusion indistinguishable from normal work. |
## 2. Adversaries
| ID | Adversary | Capability assumed | Not assumed |
| -- | --------- | ------------------ | ----------- |
| ADV1 | **Compromised LLM client** | Full control of one MCP client. Issues arbitrary tool calls, in any order, with any arguments, at machine speed. Sees every tool result. | Cannot read the operator's disk except through tools; cannot execute arbitrary local code outside the tool surface. |
| ADV2 | **Prompt injection** via repository content | Controls text the model reads and treats as instruction — issue bodies, PR descriptions, review comments, commit messages, file contents. Reaches the model on any read of untrusted content. | Holds no credential and issues no call directly. Its entire power is causing an *authorized* client to act. |
| ADV3 | **Malicious tool arguments** | Supplies hostile values to any parameter — paths, branch names, session identifiers, worktree paths, issue numbers — including traversal, injection, and confusion between look-alike identifiers. | Cannot bypass a gate that actually validates its input. |
| ADV4 | **Network attacker** | Observes and modifies traffic between client, server, and Gitea. Attempts downgrade, replay, and endpoint impersonation. | Does not hold a valid credential at the start. |
| ADV5 | **Curious operator** | Legitimate local access to the workstation: process table, `/tmp`, home directory, keychain prompts. Not malicious, but not authorized for every role either. | Does not defeat the OS keychain's own access control without a prompt. |
ADV2 is the adversary this architecture most under-models. Every other adversary must first
obtain something. Prompt injection obtains nothing: it borrows authority the client already
holds and is indistinguishable at the tool boundary from legitimate work. Each boundary
below therefore states whether it constrains ADV2 at all — and most do not, because they
authenticate the *caller*, not the *intent*.
## 3. Trust boundaries
"Crossing requires today" is what the code actually enforces at
`a143cd065ba06e1a2bdc5143a19ec156e53650ef`, not what the design intends.
| ID | Boundary | Protects | Crossing requires today | Crossing must require remotely |
| -- | -------- | -------- | ----------------------- | ------------------------------ |
| B1 | LLM client ↔ MCP server session | A1, A3, A10 — that a mutating session was established through the sanctioned client path | A single configured bind (`gitea_mcp_server.py:24728`) validated against one closed allowlist (`mcp_daemon_guard.py:49`, `mcp_daemon_guard.py:195`) — since #931 the identifier comes from deployment configuration and defaults to the local transport, so the boundary no longer rests on a literal, but it still rests on the *bind* rather than on an authenticated caller; client-managed provenance (`gitea_mcp_server.py:15415`) or a refusal (`gitea_mcp_server.py:15453`); production transport before recovery-authorization mint (`irrecoverable_provenance.py:497`, consumed at `gitea_mcp_server.py:9132` and `gitea_mcp_server.py:9381`) | An authenticated handshake issuing a server-side session identity bound to a principal, with the transport recorded in provenance. The physical proof (a pipe) must become a cryptographic one. |
| B2 | Role ↔ role | A9 — that author, reviewer, merger, and reconciler are distinct authorities | **The process boundary only.** The role is a property of the process, read once from `GITEA_MCP_PROFILE` (`gitea_config.py:54`). A caller gets author permissions by connecting to the author process. Review and merge are the operations singled out for extra care (`gitea_config.py:97`) | A per-request principal, so the role follows from the credential presented and cannot be selected by reaching a different endpoint. |
| B3 | MCP server ↔ credential store | A3, A8 — that only sanctioned code turns a profile into a token | `_keychain_token` shelling out to the login keychain (`gitea_config.py:956`), dispatched by `resolve_token` (`gitea_config.py:974`) with the reference type built at `gitea_config.py:1015`, gated by `assert_keychain_access_allowed` (`mcp_daemon_guard.py:583`). Inline secrets are rejected at config load (`gitea_config.py:294`) | A credential provider keyed by the *request* principal, returning only that principal's credential, with the source recorded and the value never returned. |
| B4 | MCP server ↔ Gitea | A1, A2 — that only authorized calls reach the forge | A bearer token over TLS. Server-side, nothing distinguishes one role's token from another beyond the account it belongs to | Unchanged at the forge; the endpoint in front of it must refuse unauthenticated and plaintext connections before tool dispatch. |
| B5 | MCP server ↔ caller's filesystem | A7 — that a tool acts on the *caller's* disk or refuses | Nothing. The server's disk *is* the caller's disk. Worktree bootstrap writes directly (`gitea_mcp_server.py:10897`); the active workspace is process-global (`gitea_mcp_server.py:193`, `gitea_mcp_server.py:194`) | An explicit per-tool classification, enforced at dispatch, refusing filesystem tools over a transport that cannot reach the caller's disk. A green verdict about the wrong disk is the failure to prevent. |
| B6 | MCP server ↔ coordination state | A6, A9 — mutual exclusion | Local files and a local SQLite database, with liveness judged from the local process table (`issue_lock_store.py:98`), keyed on paths under one user's home (`issue_lock_store.py:26`, `mcp_session_state.py:27`, `control_plane_db.py:47`) and on `os.getpid()` (`control_plane_db.py:1145`, `gitea_mcp_server.py:12804`). A legacy global slot still exists at `gitea_mcp_server.py:2351`, and the session-pointer file is named per PID (`issue_lock_store.py:83`) | One authority per ownership question, with liveness from session identity and expiry, and atomic acquire, renew, and release across hosts. |
| B7 | Gitea integration ↔ unrelated integrations | A4, A5 — that a Gitea compromise is not a CI and observability compromise | **Nothing.** See §5. The Gitea server reads Jenkins and GlitchTip secrets (`gitea_config.py:851`, reached from `gitea_config.py:837`) and holds the Sentry token (`sentry_incident_bridge.py:190`) | A hard process boundary. This is the boundary #956 exists to create. |
| B8 | Tenant ↔ tenant (`prgs` / `mdcps` / `local-lab`) | A2 — that one organization's compromise is not another's | Convention. One configuration declares all three contexts; `resolve_service` fails closed on a *disabled* context (`gitea_config.py:704`) but the credentials of enabled ones remain reachable in-process. A per-profile repository scope exists (`gitea_config.py:499`) | Separate deployments, or at minimum per-tenant credential scopes with no process able to resolve both. |
| B9 | Deployed code ↔ merged policy | A1, A10 — that the running server enforces the rules that were actually merged | Comparing this process's startup commit against this disk (`master_parity_gate.py:168`), conjoined into a single verdict (`master_parity_gate.py:255`) published by `gitea_mcp_server.py:19105` | Freshness defined against the deployed build identity, with an explicit fail-closed verdict when undeterminable. |
### What no boundary constrains
None of B1B9 constrains **ADV2**. Every one authenticates a caller or a process; prompt
injection supplies neither. An injected instruction that reaches an authorized author
session crosses B1, B2, B3, and B5 legitimately, because at each of those boundaries it *is*
the author. The only controls that bite ADV2 are those constraining what an authenticated
principal may do regardless of what it asks for — the per-role permission split (B2), the
repository scope at `gitea_config.py:499`, and separation of duty (A9). Sizing those
controls correctly matters more after the migration, not less, because a remote endpoint
raises the number of clients that can be injected into.
## 4. Data flows
Flows that cross a boundary. `==>` carries a credential; `-->` does not.
```
B1 B4
[LLM client] ====================> [MCP server] ========> [Gitea]
^ stdio pipe today | ^ (A1,A2)
| session identity | |
| after migration | |
| | | B3
untrusted repository content | +======> [macOS login keychain] (A8)
read back into the model (ADV2) | resolves A3, A4, A5
^ |
+----------------------------------+
|
B5 | B6
[operator checkout / worktrees] <--------+-------> [locks · leases · sqlite]
(A7) | (A6)
|
B7 <-- boundary does not exist today
|
+========================+========================+
| | |
[Jenkins] (A4) [GlitchTip] (A5) [Sentry] (A5)
external MCP server external MCP server in-process bridge
```
Two flows deserve attention because neither is obvious from the code:
1. **The keychain flow fans out.** B3 is drawn once but resolves credentials for *every*
configured profile and service, not only the active one. `gitea_list_profiles`
(`gitea_mcp_server.py:19261`) reports each profile's credential status by calling
`resolve_token` on it (`gitea_mcp_server.py:19312`), and `gitea_audit_config`
(`gitea_mcp_server.py:19555`) reports service credential status through
`service_summaries` (`gitea_mcp_server.py:19577`).
2. **The return path is a flow too.** Content read from Gitea travels back into the model
and is treated as instruction. This is the ADV2 edge, and it is the only edge in the
diagram with no authentication on it, because it is not a request.
## 5. Per-boundary credential inventory
**14 credentials in total.** Blast radius is stated as what the credential yields *on its
own*, assuming every gate not backed by the credential itself has been bypassed — because
an attacker holding a token calls the API, not our tools.
| ID | Credential | Holder | Boundary | Blast radius |
| -- | ---------- | ------ | -------- | ------------ |
| CR1 | `prgs-author` Gitea token — account `jcwalker3` | macOS keychain; resolved in-process (`gitea_config.py:974`) | B3 → B4 | Create branches, push, commit, open PRs, create/close/comment issues on the control-plane repo. Cannot approve or merge. The one credential whose identity is genuinely distinct. |
| CR2 | `prgs-reviewer` Gitea token — account `sysadmin` | macOS keychain | B3 → B4 | Approve and request changes. **Shares one Gitea account with CR3, CR4, CR5.** |
| CR3 | `prgs-merger` Gitea token — account `sysadmin` | macOS keychain | B3 → B4 | Merge to `master` — A1 in full. Same account as CR2. |
| CR4 | `prgs-reconciler` Gitea token — account `sysadmin` | macOS keychain | B3 → B4 | Close PRs, delete branches, irrecoverable decision-lock recovery. Same account as CR2. |
| CR5 | `prgs-controller` Gitea token — account `sysadmin` | macOS keychain | B3 → B4 | Same operation set as CR4. Same account as CR2. |
| CR6 | `mdcps-author` Gitea token — account `913443` | macOS keychain | B3 → B4, B8 | Author operations on a second organization. **Shares one account with CR7 and CR8.** |
| CR7 | `mdcps-reviewer` Gitea token — account `913443` | macOS keychain | B3 → B4, B8 | Approve and request changes on `mdcps`. Same account as CR6. |
| CR8 | `mdcps-merger` Gitea token — account `913443` | macOS keychain | B3 → B4, B8 | Merge on `mdcps` — A2 in full. Same account as CR6. |
| CR9 | MDCPS Jenkins read credential | macOS keychain, read from the Gitea server process (`gitea_config.py:851`) | B7 | Read CI jobs, builds, and logs (A4). Enabled today. |
| CR10 | MDCPS GlitchTip read credential | macOS keychain, read from the Gitea server process (`gitea_config.py:851`) | B7 | Read error events and their payloads (A5). Enabled today. |
| CR11 | `SENTRY_AUTH_TOKEN` | Process environment, read in-process (`sentry_incident_bridge.py:36`, `sentry_incident_bridge.py:190`), sent as a bearer header (`sentry_incident_bridge.py:289`) | B7 | Read and reconcile Sentry issues (A5). Not a keychain credential — an env var, so it is inherited by anything the process spawns. |
| CR12 | `SENTRY_DSN` | Process environment (`sentry_observability.py:55`) | B7 | Write events into the observability project. Low read value, real forgery value: an attacker can inject fabricated events into the record (A10). |
| CR13 | macOS login keychain access | The operator's login session; gated by `assert_keychain_access_allowed` (`mcp_daemon_guard.py:583`) | B3, ADV5 | **Every other credential in this table except CR11 and CR12.** This is the aggregation point. |
| CR14 | Coordination-store access (no secret) | Filesystem permissions — `control_plane_db.py:47`, created `0o700` (`control_plane_db.py:380`), opened with a local file lock (`control_plane_db.py:386`) | B6, ADV5 | Full read/write of locks, leases, and decision records (A6). **There is no credential here at all** — anything running as the operator can rewrite ownership. |
### Findings
**Finding 1 — Role separation is not credential separation.** Four `prgs` roles resolve to
one Gitea account (`sysadmin`): reviewer, merger, reconciler, and controller. A stolen
reviewer credential *is* a merger credential. A9 — separation of duty between approving and
landing — is therefore enforced entirely by which local process a call reaches (B2), and not
at all by the forge. It survives exactly as long as B2 does, and B2 is the boundary the
migration dissolves.
**Finding 2 — The `mdcps` tenant has no role separation at all.** Author, reviewer, and
merger all resolve to account `913443`. One credential can open a PR, approve it, and merge
it. The in-process self-review check compares the authenticated username against the PR
author and would refuse — but that check runs on our side of B4. It is not a property of
the credential, and an attacker holding the token does not call our tools.
**Finding 3 — Any one role process can resolve every other role's credential.** This is not
inferred; it is demonstrated by tool output. `gitea_list_profiles`
(`gitea_mcp_server.py:19261`) called from the **author** session reports
`identity_status: "credentials present"` for `prgs-merger`, `prgs-reviewer`,
`prgs-reconciler`, and every `mdcps` profile, because it calls `resolve_token` on each one
(`gitea_mcp_server.py:19312`). The author process does not merely *have access to* the
merger's credential — it reads it to answer a status query. B2 is not a credential boundary
in either direction.
**Finding 4 — The Gitea server reads CI and observability secrets.** `gitea_audit_config`
(`gitea_mcp_server.py:19555`) reports `MDCPS Jenkins: enabled, read-only, authenticated`.
That word `authenticated` is produced by `service_summaries` (`gitea_mcp_server.py:19577`,
defined at `gitea_config.py:837`), whose default check calls `_keychain_token` on the
service's own keychain reference (`gitea_config.py:851`). Producing that one line requires
the Gitea MCP server to read the Jenkins secret and the GlitchTip secret out of the
keychain. B7 does not exist.
**Finding 5 — Jenkins and GlitchTip are already decomposed; the reach is residual.** Their
tools live in separately registered servers, marked `external-mcp`
(`gitea_mcp_server.py:17710`, `gitea_mcp_server.py:17716`, `gitea_mcp_server.py:17737`,
`gitea_mcp_server.py:17742`) with their own expected tool sets (`mcp_discoverability.py:9`,
`mcp_discoverability.py:17`). The correct decomposition was already chosen. What remains is
a leak across it: the credential *references* still live in the Gitea configuration and are
still resolved by the Gitea process. #75 bundled these services into one control-plane
umbrella; the tools were separated afterwards, the credentials were not.
**Finding 6 — Sentry is the exception that is not decomposed.** Unlike Jenkins and
GlitchTip, the Sentry bridge runs *inside* the Gitea server, resolving its token from the
process environment (`sentry_incident_bridge.py:190`) and sending it as a bearer header
(`sentry_incident_bridge.py:289`). Being an environment variable rather than a keychain item
makes it strictly worse: it needs no keychain prompt and is inherited by every subprocess the
server spawns — including the `ps` invocations at `gitea_mcp_server.py:21465` and
`gitea_mcp_server.py:21509`, reached from `gitea_mcp_server.py:21445`.
**Finding 7 — The highest-value coordination asset has the weakest gate.** A6 is protected
by filesystem permissions alone (CR14). Corrupting a lease requires no Gitea credential,
produces no forge-side audit record, and breaks the mutual exclusion the entire workflow
assumes. Every other asset costs an attacker a credential; this one costs nothing beyond
local access, which is exactly ADV5's position.
**Finding 8 — Provenance authenticates the launch, not the caller.** `server_provenance` is
reported as exactly `client_managed` or `manual_launch` (`gitea_mcp_server.py:19004`),
derived from environment inspection (`gitea_mcp_server.py:15415`) with the recognized-key
allowlist at `gitea_config.py:1172` and the generator that emits the marker at
`gitea_config.py:1233`. Every one of those facts is fixed at process start. A client that is
trustworthy at launch and compromised a minute later remains `client_managed` for the life
of the process, and the transport contract that underwrites it is stated as a property of
the server itself (`mcp_server.py:4`). Since #931 that contract names the configured
transport rather than asserting stdio, but it is still fixed once, at bind, for the life of
the process.
## 6. Decomposition ruling
This section is the ruling #956 requires. It is a decision, not a recommendation.
**D1 — No unrelated co-residency.** A single integration process **must not** hold, resolve,
or be able to resolve credentials for services it does not itself integrate with.
Concretely: the Gitea MCP service may hold Gitea credentials and nothing else. Jenkins,
GlitchTip, Sentry, and any database credential are **not permitted** to co-reside with Gitea
credentials in one process.
*Rationale.* A process is the smallest unit an attacker takes whole. Once ADV1 or ADV2
controls execution in a process, every credential that process can resolve is theirs, and no
in-process check helps, because the checks are in the process too. Blast radius is therefore
a property of the process boundary and nothing finer. Findings 4 and 6 show that today one
compromise of the Gitea server yields CI read access, error-tracking read access, and — via
CR13 — every role credential on both tenants. That is the single largest reduction in blast
radius available anywhere in epic #929, and it costs no new mechanism: the decomposition
already exists (Finding 5) and is merely leaked across.
**D2 — Separation of duty must be backed by credentials.** Two roles whose separation is a
security property must not resolve to the same forge account. Specifically, reviewer and
merger must be distinct accounts. Today they are not, on either tenant (Findings 1 and 2).
*Rationale.* B2 is a process boundary, and the migration's entire purpose is to replace
process boundaries with request-level ones. A separation enforced only by which process a
call reaches does not survive that replacement — and it is already bypassable by anyone who
holds the token and calls the API instead of the tool.
**D3 — Credential resolution is scoped to the request principal.** A session must resolve its
own credential and must have no path to any other principal's. The resolve-every-profile
behavior behind `gitea_mcp_server.py:19312` and `gitea_mcp_server.py:19577` must report
configured-or-not from configuration alone, without resolving the secret.
*Rationale.* Finding 3. An audit surface that proves a credential exists by fetching it is a
credential-aggregation primitive wearing a diagnostic's clothes.
**D4 — Coordination state is a protected asset with its own authority.** Access to locks,
leases, and decision records must require an authenticated session, not merely local
filesystem access.
*Rationale.* Finding 7. #937 already moves this store for concurrency reasons; the
authorization requirement must land with it, or the store becomes remotely reachable while
still being authorized by nothing.
### Exceptions
**One, time-boxed.** During the dual-run window defined by #939, the **local** stdio fleet
may continue to resolve Jenkins and GlitchTip credential *references* from the shared
configuration, because removing them from the local configuration is not a prerequisite for
standing up the remote endpoint and would strand the operator's existing local workflow.
This exception is bounded by all of:
- It applies to the local stdio deployment only. The remote endpoint (#938) must be
configured with Gitea credentials and no others from its first day.
- It expires when #939 completes. It does not survive cutover.
- It does not extend to Sentry: CR11 and CR12 are process-environment credentials in the
Gitea server (Finding 6) and must be absent from the remote deployment's environment
regardless of dual-run state.
No exception is granted to D2, D3, or D4.
### Consequences for the target architecture
- The remote endpoint serves **Gitea only**. It is not a general control-plane endpoint.
- Jenkins and GlitchTip keep their existing separate servers, and their credential
references move out of the Gitea configuration.
- The Sentry bridge either moves behind its own service boundary or is absent from the
remote deployment. It does not travel with the Gitea server.
- Reviewer and merger accounts diverge before the endpoint is trusted for merges, or A9 is
recorded as unenforced.
## 7. Child-to-boundary mapping
Every #929 child from 2 through 10, mapped to the boundary it implements. A child
implementing more than one boundary names its primary first.
| Child | Issue | Boundaries | What it must establish | Rulings it must honor |
| ----: | ----- | ---------- | ---------------------- | --------------------- |
| 2 | #931 | B1, B9 | The bound transport becomes a validated value that provenance and freshness can both key on. Without it neither B1 nor B9 has an input. | — |
| 3 | #932 | B2 | The role becomes a property of the request, not the process — the boundary the migration otherwise deletes. | D2, D3 |
| 4 | #933 | B3, B7 | Credentials come from a provider keyed by principal. This is where D1 and D3 are either enforced or permanently lost. | D1, D3 |
| 5 | #934 | B1 | Session provenance replaces pipe-and-process-table proof with an authenticated session identity. | — |
| 6 | #935 | B9 | Freshness redefined against deployed build identity, with an explicit undeterminable verdict. | — |
| 7 | #936 | B5 | Every tool classified and the filesystem boundary enforced at dispatch, so a tool cannot return green about the wrong disk. | — |
| 8 | #937 | B6 | One authority per ownership question, with session-identity liveness and atomic transitions. | D4 |
| 9 | #938 | B4, B1, B8 | The endpoint: authentication, principal binding, transport security, and — critically — the deployed credential set. | D1, D2, D3 |
| 10 | #939 | B6 | Dual-run with exactly one coordination authority at every instant, and the rollback that proves the way back. | D1 exception expiry |
Boundary coverage: B1 (#931, #934, #938), B2 (#932), B3 (#933), B4 (#938), B5 (#936),
B6 (#937, #939), B7 (#933), B8 (#938), B9 (#931, #935).
B7 has exactly one owner, #933, and that is deliberate. B7 is not created by standing up an
endpoint; it is created by deciding which credentials a process may resolve, which is
precisely what the credential-provider child does.
## 8. Adversarial walkthrough: a fully compromised LLM client
The required negative case. ADV1 holds one MCP client and issues arbitrary tool calls. What
does it reach at each boundary?
### Before the migration
The client is attached to the local fleet over stdio.
| Boundary | What ADV1 reaches | Stopped by |
| -------- | ----------------- | ---------- |
| B1 | Everything the fleet serves. The client *is* the sanctioned launcher: it satisfies the client-managed check (`gitea_mcp_server.py:15415`) by construction, and provenance is never re-verified after launch (Finding 8). | Nothing. The guard authenticates the launch, not the caller. |
| B2 | All five roles — it is attached to all five namespaces. It can author a PR, approve it from the reviewer namespace, and merge it from the merger namespace. | Only the in-process self-review check, which compares `jcwalker3` (author) against `sysadmin` (reviewer) and **passes**, because Finding 1 made them different accounts while leaving reviewer and merger identical. A9 falls in one sequence of legitimate calls. |
| B3 | Every credential in CR1CR10 via CR13, with no additional prompt — the daemon is already sanctioned, so `assert_keychain_access_allowed` (`mcp_daemon_guard.py:583`) returns immediately. | Nothing. |
| B4 | A1 and A2 in full. | Branch protection at the forge, to the extent configured. |
| B5 | The operator's checkout and every worktree, through the author tools (`gitea_mcp_server.py:10897`), plus the shared stderr path at `mcp_server.py:13`. | Nothing; the server's disk is the target disk. |
| B6 | All coordination state — no credential required (CR14). It can forge lease ownership and clear decision locks. | Filesystem permissions, which it already satisfies. |
| B7 | Jenkins (A4) and GlitchTip (A5) secrets via Finding 4, and CR11/CR12 from its own environment. | Nothing. |
| B8 | Both tenants. | Nothing in-process; only the disabled-context check (`gitea_config.py:704`), which does not apply to enabled contexts. |
| B9 | Can it run stale code? Only by advancing the checkout, which parity detects (`master_parity_gate.py:255`). | This gate holds — the one boundary that behaves as designed against ADV1. |
**Result: total.** One compromised client yields A1 through A8 and A10. The only asset with
real resistance is A1 via branch protection, and the client holds the merger credential
anyway. Nine boundaries, one meaningful stop.
### After the migration
The same client authenticates to the remote endpoint with one role's credential, assuming
#931#939 land **and honor D1D4**.
| Boundary | What ADV1 reaches | Stopped by |
| -------- | ----------------- | ---------- |
| B1 | One authenticated session, bound to one principal. | #934: a forged or expired session identity is refused; the client cannot mint one. |
| B2 | **One role.** Presenting the author credential yields author permissions only. | #932: the principal comes from the credential, not from which endpoint was reached. |
| B3 | **One credential — its own.** | #933 with D3: the provider resolves by principal, and no diagnostic resolves the others. |
| B4 | That role's authority on the forge. | Endpoint authentication (#938); plaintext and unauthenticated attempts refused before dispatch. |
| B5 | **Nothing.** Filesystem tools are refused over the remote transport with a named blocker. | #936. |
| B6 | Its own leases; contention resolves to exactly one winner. | #937 with D4: authenticated session required, not filesystem access. |
| B7 | **Nothing.** No CI or observability credential exists in the process. | D1 — the single largest reduction on this table. |
| B8 | One tenant. | D1 and #938: the deployment carries one tenant's credentials. |
| B9 | Cannot induce stale enforcement. | #935: explicit fail-closed verdict, including undeterminable. |
**Result: bounded.** The compromise is contained to one role on one tenant, with no
filesystem reach and no lateral credential access. A9 survives *only if D2 lands* — if
reviewer and merger still share `sysadmin`, a compromised reviewer session still merges, and
this row reads the same after the migration as before it.
### What the migration does not fix
Against **ADV2**, both tables are identical. Prompt injection does not need to cross a
boundary: it arrives inside an authorized session and asks that session to do what it is
already permitted to do. Every "stopped by" above authenticates a principal, and the
injected instruction has the correct principal. The migration reduces ADV1's blast radius by
roughly an order of magnitude and reduces ADV2's by nothing.
The controls that do constrain ADV2 are per-principal permission scope (#932), repository
scope (`gitea_config.py:499`), and credential-backed separation of duty (D2) — each limiting
what an authenticated session may do *regardless of what it is asked for*. #955's
secure-isolation end state should be read with that distinction in mind: removing credentials
from clients defeats ADV1 and ADV5, and does not by itself defeat ADV2.
Two further items are explicitly out of scope here and unowned by #929:
- **Session-credential rotation and revocation.** #938 names rotation as documentation, but
no child owns proving that a revoked credential stops an in-flight session.
- **ADV3** (malicious tool arguments) is diffused across every child rather than owned. The
per-request principal work in #932 is the natural place to assert that identifiers taken
from the request never authorize anything on their own.
## 9. How to verify this document
1. `PYTHONPATH=. pytest tests/test_issue_956_threat_model.py` — resolves every anchor
against the working tree and checks the document's structural obligations.
2. Pick any five anchors at random and read them; the fixture states what each line must
contain.
3. Reproduce Findings 3 and 4 live: call `gitea_list_profiles` and `gitea_audit_config`
from the **author** namespace. Credential presence reported for roles other than the
active one is Finding 3; `MDCPS Jenkins: enabled, read-only, authenticated` is Finding 4.
If the anchor test fails after an unrelated refactor, the anchors moved and the fixture
needs regenerating — the claims are still true, but they are no longer traceable, which
#956 treats as the same defect.
+17 -3
View File
@@ -1,7 +1,10 @@
#!/usr/bin/env python3
"""Gitea MCP Server — exposes Gitea operations as MCP tools.
Runs over stdio. All tools authenticate via macOS keychain (git credential fill).
The transport is selected by deployment configuration (GITEA_MCP_TRANSPORT) and
defaults to the local client-spawned transport when unset (#931); the permitted
set lives in mcp_transport_config. All tools authenticate via macOS keychain
(git credential fill).
Usage (standalone test):
python3 mcp_server.py
@@ -24718,7 +24721,11 @@ if __name__ == "__main__":
# basename-only stack frames, and import-only launch cannot reconstruct
# native transport; offline imports / standalone scripts fail closed.
mcp_daemon_guard.mark_sanctioned_daemon()
mcp_daemon_guard.bind_native_mcp_transport(transport="stdio")
# #931: the transport is no longer a literal here. It comes from deployment
# configuration (GITEA_MCP_TRANSPORT), defaults to stdio when unset, and is
# validated against the single permitted set in mcp_transport_config. An
# unregistered identifier raises here, before any tool can dispatch.
mcp_daemon_guard.bind_native_mcp_transport()
# Lock this session's launch profile into the environment so child CLI
# processes (e.g. review_pr.py) can detect and refuse profile
# side-channel overrides (#199).
@@ -24736,4 +24743,11 @@ if __name__ == "__main__":
sys.stderr.write(
f"--- Sentry observability: {_sentry_status.get('reason')} ---\n"
)
mcp.run(transport="stdio")
# #931: serve over exactly the transport that was bound and pinned, and only
# if this entrypoint is commissioned to execute it. Recognition is not
# execution authorization (review 635): a registered remote identifier is
# bound, pinned and recorded, but serving it would start a listener with no
# authentication or per-request principal, which #938 owns. An absent bind
# keeps its pre-existing #695 failure; an uncommissioned transport fails
# closed here, before any listener exists and before any tool can dispatch.
mcp.run(transport=mcp_daemon_guard.authorize_transport_execution("tool service"))
+6
View File
@@ -500,6 +500,11 @@ def assess_transport_for_auth_mint() -> dict[str, Any]:
native = mcp_daemon_guard.is_native_mcp_transport()
pytest = mcp_daemon_guard.is_pytest_runtime()
production = mcp_daemon_guard.is_production_native_mcp_transport()
# #931: report which transport underwrites the verdict, read through the
# one shared accessor rather than assumed to be stdio. The gate's decision
# is unchanged here; naming the transport is what lets #932 re-derive the
# guarantee from an authenticated session instead of from the bind.
bound = mcp_daemon_guard.bound_transport()
if not native and not pytest:
reasons.append(
"irrecoverable provenance authorization requires production native "
@@ -510,6 +515,7 @@ def assess_transport_for_auth_mint() -> dict[str, Any]:
"allowed": not reasons,
"native_mcp_transport": native,
"production_native_mcp_transport": production,
"bound_transport": bound,
"pytest": pytest,
"reasons": reasons,
}
+176 -13
View File
@@ -32,6 +32,8 @@ import time
from pathlib import Path
from typing import Any
import mcp_transport_config
SANCTIONED_DAEMON_ENV = "GITEA_MCP_SANCTIONED_DAEMON"
ALLOW_DIRECT_IMPORT_ENV = "GITEA_ALLOW_DIRECT_MCP_IMPORT"
ALLOW_KEYCHAIN_CLI_ENV = "GITEA_ALLOW_KEYCHAIN_CLI"
@@ -42,7 +44,9 @@ FORCE_PROVENANCE_FAIL_ENV = "GITEA_TEST_FORCE_UNSANCTIONED"
_NATIVE_RUNTIME: dict[str, Any] | None = None
# Production transport identifiers accepted by bind_native_mcp_transport.
_PRODUCTION_TRANSPORTS = frozenset({"stdio"})
# #931: the permitted set is defined once, in mcp_transport_config. This name
# is kept as an alias so the guard never restates a transport identifier.
_PRODUCTION_TRANSPORTS = mcp_transport_config.SUPPORTED_TRANSPORTS
_RUNTIME_MODE_PRODUCTION = "production"
_RUNTIME_MODE_TEST = "test"
_PHASE_ENTRYPOINT_CLAIMED = "entrypoint_claimed"
@@ -58,6 +62,23 @@ class UnsanctionedRuntimeError(RuntimeError):
"""Raised when mutation/credential code runs outside a native MCP daemon."""
class TransportExecutionError(UnsanctionedRuntimeError):
"""Raised when a bound transport may not be served by this entrypoint (#931).
Subclasses :class:`UnsanctionedRuntimeError` so every existing fail-closed
handler still catches it, while letting a caller that cares distinguish
"nothing is bound" from "something valid is bound but its listener has not
been commissioned". Carries the structured verdict on ``.assessment``.
"""
def __init__(self, message: str, assessment: dict[str, Any] | None = None):
super().__init__(message)
self.assessment = assessment or {}
self.blocker_kind = self.assessment.get("blocker_kind")
self.owner_issue = self.assessment.get("owner_issue")
self.transport = self.assessment.get("transport")
def is_pytest_runtime() -> bool:
if (os.environ.get(FORCE_PROVENANCE_FAIL_ENV) or "").strip() in {
"1",
@@ -171,23 +192,46 @@ def mark_sanctioned_daemon() -> dict[str, Any]:
return native_runtime_status()
def bind_native_mcp_transport(*, transport: str) -> dict[str, Any]:
"""Bind the live native MCP transport lifecycle (#695).
def bind_native_mcp_transport(*, transport: str | None = None) -> dict[str, Any]:
"""Bind the live native MCP transport lifecycle (#695 / #931).
Must be called from the resolved canonical entrypoint immediately before
the real MCP server transport loop (e.g. ``mcp.run(transport=\"stdio\")``).
Requires a prior successful :func:`mark_sanctioned_daemon` claim in this
process. Import-only or offline launch without this bind leaves
the real MCP server transport loop (``mcp.run``). Requires a prior
successful :func:`mark_sanctioned_daemon` claim in this process.
Import-only or offline launch without this bind leaves
:func:`is_native_mcp_transport` false.
#931: ``transport`` is now optional. Omitting it — which is what the
production entrypoint does — resolves the identifier from deployment
configuration via :func:`mcp_transport_config.resolve_configured_transport`,
yielding :data:`mcp_transport_config.DEFAULT_TRANSPORT` when nothing is
configured. An explicit argument remains supported for tests and for a
launcher that has already resolved the value. Either way the identifier is
validated against the single permitted set before the runtime record is
written, so no tool can dispatch over an unregistered transport.
The resolved value is pinned into the process-local record and is read back
only through :func:`bound_transport`. Rebinding to a different transport is
refused, so two guards can never observe different values in one process.
"""
global _NATIVE_RUNTIME
transport_name = (transport or "").strip().lower()
if transport_name not in _PRODUCTION_TRANSPORTS:
raise UnsanctionedRuntimeError(
f"bind_native_mcp_transport rejected: transport {transport!r} is "
f"not a production MCP transport (#695). Allowed: "
f"{sorted(_PRODUCTION_TRANSPORTS)}."
)
if transport is None:
resolution = mcp_transport_config.resolve_configured_transport()
transport_name = str(resolution["transport"])
if not resolution["supported"]:
raise UnsanctionedRuntimeError(
"bind_native_mcp_transport rejected: "
+ "; ".join(resolution["reasons"])
+ " No tool is served over an unregistered transport."
)
else:
transport_name = mcp_transport_config.normalize_transport(transport)
if transport_name not in _PRODUCTION_TRANSPORTS:
raise UnsanctionedRuntimeError(
f"bind_native_mcp_transport rejected: transport {transport!r} is "
f"not a production MCP transport (#695). Allowed: "
f"{sorted(_PRODUCTION_TRANSPORTS)}."
)
entrypoint_path = _caller_official_entrypoint_path()
if entrypoint_path is None:
@@ -216,6 +260,23 @@ def bind_native_mcp_transport(*, transport: str) -> dict[str, Any]:
"between mark and bind (#695)."
)
# #931: one process binds one transport. Re-binding the same identifier is
# idempotent (a retried launch step must not fail); re-binding a different
# one is refused, because a guard that already read the first value would
# otherwise disagree with a guard that reads the second.
already_bound = (_NATIVE_RUNTIME.get("transport") or "").strip()
if (
already_bound
and _NATIVE_RUNTIME.get("phase") == _PHASE_TRANSPORT_BOUND
and already_bound != transport_name
):
raise UnsanctionedRuntimeError(
"bind_native_mcp_transport rejected: transport is already bound to "
f"{already_bound!r} in this process; rebinding to "
f"{transport_name!r} is forbidden (#931). Restart the server to "
"change the deployment transport."
)
# Pin session-state root for this server lifetime (#695 AC2 / PR #701).
# Changing GITEA_MCP_SESSION_STATE_DIR after bind must not manufacture a
# second authority domain for decision locks / workflow proofs.
@@ -353,6 +414,88 @@ def is_production_native_mcp_transport() -> bool:
return (_NATIVE_RUNTIME or {}).get("mode") == _RUNTIME_MODE_PRODUCTION
def bound_transport() -> str | None:
"""The one authoritative bound transport identifier, or ``None`` (#931).
This is the shared accessor every transport-aware guard reads. It reports
the value pinned at bind time, never the environment, so changing
``GITEA_MCP_TRANSPORT`` after the bind cannot move what a guard observes —
the same rule :func:`pinned_session_state_dir` applies to session state.
``None`` means unbound: an offline import or a launch that never reached
the bind. Callers must treat that as fail-closed, exactly as they already
treat :func:`is_native_mcp_transport` returning false.
"""
if not is_native_mcp_transport():
return None
return (_NATIVE_RUNTIME or {}).get("transport") or None
def assert_transport_bound(context: str = "tool service") -> str:
"""Return the bound transport, or fail closed before *context* (#931).
Called immediately before the server enters its transport loop so an
invalid or absent bind stops the process rather than serving tools over a
transport no guard can name.
"""
transport = bound_transport()
if transport:
return transport
raise UnsanctionedRuntimeError(
f"No MCP transport is bound; refusing {context} (#931). "
"bind_native_mcp_transport must succeed from the canonical entrypoint "
"before any tool is served. Offline import and standalone launch "
"cannot reconstruct a bind."
)
def assess_serve_authorization() -> dict[str, Any]:
"""Structured verdict on whether this process may serve tools (#931).
This is the decision that consumes :func:`bound_transport`. It is what stops
the bound identifier from being reporting-only metadata: the serve path
cannot proceed unless the value pinned at bind is one this entrypoint is
commissioned to execute.
Never raises; returns the verdict so callers and diagnostics can inspect it.
"""
return mcp_transport_config.assess_transport_execution(bound_transport())
def authorize_transport_execution(context: str = "tool service") -> str:
"""Return the transport this process may serve, or fail closed (#931).
Two distinct boundaries, in order:
1. **Bind presence** — :func:`assert_transport_bound` enforces the
pre-existing #695 contract, so an unbound runtime keeps its established
failure and reason code.
2. **Execution authorization** — the bound identifier must be one this
entrypoint is commissioned to serve. A registered transport whose
listener has not been commissioned is refused here, before any listener
is created and before any tool can dispatch.
That ordering matters: recognition, validation and durable recording all
still happen for a remote identifier, so #931's seam is intact; only the act
of *serving* it is withheld until its owning issue commissions it.
"""
# Boundary 1: unbound stays exactly as fail-closed as it was under #695.
assert_transport_bound(context)
# Boundary 2: bound, but is this entrypoint allowed to serve it?
assessment = assess_serve_authorization()
if assessment.get("allowed"):
return str(assessment["transport"])
reasons = "; ".join(assessment.get("reasons") or []) or "not authorized"
next_action = assessment.get("exact_next_action") or ""
raise TransportExecutionError(
f"Refusing {context} (#931) [{assessment.get('blocker_kind')}]: "
f"{reasons} {next_action}".strip(),
assessment,
)
def is_sanctioned_mcp_daemon() -> bool:
"""Backward-compatible name; #695 requires native transport, not env alone."""
if is_production_native_mcp_transport():
@@ -476,6 +619,21 @@ def native_runtime_status() -> dict[str, Any]:
"entrypoint_path": rt.get("entrypoint_path"),
"phase": rt.get("phase"),
"transport": rt.get("transport"),
# #931: the authoritative bound identifier, plus the seam that defines
# what may be bound. ``bound_transport`` is None until a bind succeeds,
# so an offline import is distinguishable from a stdio session.
"bound_transport": bound_transport(),
"transport_bound": bound_transport() is not None,
"default_transport": mcp_transport_config.DEFAULT_TRANSPORT,
"supported_transports": list(mcp_transport_config.supported_transports()),
"transport_env": mcp_transport_config.TRANSPORT_ENV,
# #931 review 635: recognition and execution authorization are distinct.
# ``supported`` is what may be bound; ``executable`` is what this
# entrypoint may actually serve. A recognized-but-uncommissioned
# transport reports serve_authorized False with a named blocker.
"executable_transports": list(mcp_transport_config.executable_transports()),
"serve_authorized": bool(assess_serve_authorization().get("allowed")),
"serve_authorization": assess_serve_authorization(),
"mode": rt.get("mode"),
"session_state_dir": pinned_session_state_dir() or rt.get("session_state_dir"),
"session_state_dir_pinned": pinned_session_state_dir() is not None,
@@ -502,7 +660,12 @@ def mutation_provenance_fields() -> dict[str, Any]:
if st.get("mode") == _RUNTIME_MODE_TEST and st["native_mcp_transport"]:
transport = "test_native_mcp"
return {
# ``transport`` stays the trust *class* it has always been, so existing
# durable records keep their shape. ``bound_transport`` (#931) adds the
# bound identifier itself, which is what lets an operator tell from a
# durable record which transport performed a mutation.
"transport": transport,
"bound_transport": st.get("bound_transport"),
"native_mcp_transport": bool(st["native_mcp_transport"]),
"production_native_mcp_transport": bool(
st.get("production_native_mcp_transport")
+7 -3
View File
@@ -1,7 +1,10 @@
#!/usr/bin/env python3
"""Gitea MCP Server — exposes Gitea operations as MCP tools.
Runs over stdio. All tools authenticate via macOS keychain (git credential fill).
The transport is selected by deployment configuration (GITEA_MCP_TRANSPORT) and
defaults to the local client-spawned transport when unset (#931); the permitted
set lives in mcp_transport_config. All tools authenticate via macOS keychain
(git credential fill).
"""
import os
import sys
@@ -43,8 +46,9 @@ check_conflict_markers()
# #558 / #695: claim the official entrypoint before loading mutation modules.
# This alone does NOT authorize mutations — gitea_mcp_server binds the live
# native MCP transport (stdio) immediately before mcp.run. Import-only or
# offline launch without that bind fails closed on mutations.
# native MCP transport immediately before mcp.run, over the configured
# transport (#931). Import-only or offline launch without that bind fails
# closed on mutations.
try:
import mcp_daemon_guard
+4
View File
@@ -588,9 +588,13 @@ def save_state(
bool(prov.get("production_native_mcp_transport")),
)
body.setdefault("transport", prov.get("transport"))
# #931: record the bound transport identifier itself, so a durable
# decision lock names which transport performed the mutation.
body.setdefault("bound_transport", prov.get("bound_transport"))
except Exception:
body.setdefault("native_mcp_transport", False)
body.setdefault("transport", "untrusted")
body.setdefault("bound_transport", None)
envelope = {
"kind": kind,
+292
View File
@@ -0,0 +1,292 @@
"""Single authoritative source for the bound MCP transport identifier (#931).
Before this module the transport was a literal, passed once at the bottom of
``gitea_mcp_server`` as ``bind_native_mcp_transport(transport="stdio")``. Every
guard that later asks "is this a trusted native session" resolves that question
through the value bound there, so the literal was effectively a constant in the
authorization chain rather than configuration.
This module is the seam. It owns three things and nothing else:
- the permitted set of transport identifiers,
- the default used when deployment configuration says nothing,
- the resolution of the configured value into a validated identifier.
It deliberately holds no state. The *bound* transport is pinned once, at bind
time, into the process-local native-runtime record owned by
:mod:`mcp_daemon_guard`, and is read back through
``mcp_daemon_guard.bound_transport()``. That split matters: configuration is
read exactly once, before any tool can dispatch, so a later environment change
cannot move the value a guard observes — the same pinning rule already applied
to the session-state root under #695 AC2.
Nothing here consumes tool arguments, request bodies, or provenance fields. The
only input is the deployment environment, read at bind time.
Standing up a listener for a non-stdio transport is #938; this module only
makes the identifier expressible and validated.
"""
from __future__ import annotations
import os
from typing import Any, Mapping
# Deployment configuration key. Read once, at bind time, and never again.
TRANSPORT_ENV = "GITEA_MCP_TRANSPORT"
# The local, client-spawned transport. Unset configuration resolves to this,
# which is what keeps every existing stdio deployment byte-identical.
DEFAULT_TRANSPORT = "stdio"
# The sanctioned remote transport identifier. Accepting it here is what makes
# the bind pluggable; the endpoint that serves it belongs to #938. The name
# matches the MCP transport name so no second vocabulary has to be mapped.
REMOTE_TRANSPORT = "streamable-http"
# The permitted set. This is the only place transport identifiers are
# enumerated; guards consult it rather than restating any member.
#
# ``sse`` is a real MCP transport and is deliberately absent: it is the
# superseded remote transport, and admitting it would give the deployment two
# remote paths to reason about. An unregistered identifier must fail closed at
# bind time, and ``sse`` is held to that rule like any other.
SUPPORTED_TRANSPORTS = frozenset({DEFAULT_TRANSPORT, REMOTE_TRANSPORT})
# Recognition is not execution authorization (#931, review 635 B1/B2).
#
# SUPPORTED_TRANSPORTS answers "is this an identifier this system knows, and may
# it be bound, pinned and recorded?". It deliberately includes the remote
# identifier, because #931 requires the bind to become pluggable.
#
# EXECUTABLE_TRANSPORTS answers a strictly narrower question: "is this entrypoint
# commissioned to actually *serve* on that transport?". Only the local transport
# is. Handing ``streamable-http`` to ``mcp.run`` would start FastMCP's HTTP
# listener with no authentication, no TLS and no per-request principal — the
# endpoint #938 owns and gates. Recognition must therefore never imply execution.
#
# #938 commissions the remote listener by adding REMOTE_TRANSPORT here, together
# with the authentication and principal boundary its acceptance criteria require.
EXECUTABLE_TRANSPORTS = frozenset({DEFAULT_TRANSPORT})
# Which issue owns commissioning each recognized-but-not-executable transport.
# Used to make the refusal actionable rather than a generic denial.
TRANSPORT_EXECUTION_OWNER = {REMOTE_TRANSPORT: "#938"}
BLOCKER_TRANSPORT_NOT_BOUND = "transport_not_bound"
BLOCKER_TRANSPORT_NOT_RECOGNIZED = "transport_not_recognized"
BLOCKER_LISTENER_NOT_COMMISSIONED = "transport_listener_not_commissioned"
SOURCE_CONFIGURED = "deployment_configuration"
SOURCE_DEFAULT = "default"
class TransportConfigurationError(ValueError):
"""Raised when configured transport is outside :data:`SUPPORTED_TRANSPORTS`."""
def normalize_transport(value: Any) -> str:
"""Canonical form of a transport identifier; ``""`` when there is none.
Non-string values normalize to ``""`` rather than being coerced, so a
structured object smuggled in from a caller can never match a member of the
permitted set.
"""
if not isinstance(value, str):
return ""
return value.strip().lower()
def supported_transports() -> tuple[str, ...]:
"""Permitted identifiers, sorted, for messages and status payloads."""
return tuple(sorted(SUPPORTED_TRANSPORTS))
def is_supported_transport(value: Any) -> bool:
"""True when *value* normalizes to a member of the permitted set."""
return normalize_transport(value) in SUPPORTED_TRANSPORTS
def is_remote_transport(value: Any) -> bool:
"""True when *value* is a permitted transport that is not the local one."""
name = normalize_transport(value)
return name in SUPPORTED_TRANSPORTS and name != DEFAULT_TRANSPORT
def executable_transports() -> tuple[str, ...]:
"""Transports this entrypoint is commissioned to serve, sorted."""
return tuple(sorted(EXECUTABLE_TRANSPORTS))
def is_executable_transport(value: Any) -> bool:
"""True when *value* may actually be served by this entrypoint (#931).
Strictly narrower than :func:`is_supported_transport`. A recognized
identifier that is not executable is a correct, fully-bound configuration
whose listener has simply not been commissioned yet.
"""
return normalize_transport(value) in EXECUTABLE_TRANSPORTS
def assess_transport_execution(value: Any) -> dict[str, Any]:
"""Structured serve-authorization verdict for a bound transport (#931).
This is the decision that separates a *recognized* transport from one this
entrypoint may execute. It is deliberately a pure function of the bound
identifier so the serve path cannot reach a listener the deployment has not
commissioned.
Args:
value: The bound transport identifier, or ``None`` when unbound.
Returns:
dict with ``transport``, ``recognized``, ``executable``, ``allowed``,
``blocker_kind``, ``owner_issue``, ``reasons`` and
``exact_next_action``. ``allowed`` is true only for a bound, recognized,
commissioned transport.
"""
name = normalize_transport(value)
if not name:
return {
"transport": None,
"recognized": False,
"executable": False,
"allowed": False,
"blocker_kind": BLOCKER_TRANSPORT_NOT_BOUND,
"owner_issue": None,
"supported_transports": list(supported_transports()),
"executable_transports": list(executable_transports()),
"reasons": [
"no transport is bound; the serve path is fail-closed until "
"bind_native_mcp_transport succeeds (#695/#931)"
],
"exact_next_action": (
"Launch through the canonical entrypoint so "
"bind_native_mcp_transport runs before tool service."
),
}
recognized = name in SUPPORTED_TRANSPORTS
if not recognized:
return {
"transport": name,
"recognized": False,
"executable": False,
"allowed": False,
"blocker_kind": BLOCKER_TRANSPORT_NOT_RECOGNIZED,
"owner_issue": None,
"supported_transports": list(supported_transports()),
"executable_transports": list(executable_transports()),
"reasons": [
f"transport {name!r} is not a registered MCP transport (#931); "
"it should have been refused at bind time"
],
"exact_next_action": (
f"Set {TRANSPORT_ENV} to one of {list(supported_transports())}."
),
}
if name in EXECUTABLE_TRANSPORTS:
return {
"transport": name,
"recognized": True,
"executable": True,
"allowed": True,
"blocker_kind": None,
"owner_issue": None,
"supported_transports": list(supported_transports()),
"executable_transports": list(executable_transports()),
"reasons": [],
"exact_next_action": None,
}
owner = TRANSPORT_EXECUTION_OWNER.get(name)
owner_text = owner or "the issue that commissions this transport's listener"
return {
"transport": name,
"recognized": True,
"executable": False,
"allowed": False,
"blocker_kind": BLOCKER_LISTENER_NOT_COMMISSIONED,
"owner_issue": owner,
"supported_transports": list(supported_transports()),
"executable_transports": list(executable_transports()),
"reasons": [
f"transport {name!r} is registered and was bound and recorded, but "
f"this entrypoint is not commissioned to serve it (#931). Serving it "
f"would start a listener with no authentication, no transport "
f"security and no per-request principal; that endpoint is owned by "
f"{owner_text}."
],
"exact_next_action": (
f"Serve on {DEFAULT_TRANSPORT} until {owner_text} commissions the "
f"{name!r} listener with its authentication and principal boundary, "
f"which adds {name!r} to EXECUTABLE_TRANSPORTS."
),
}
def resolve_configured_transport(
env: Mapping[str, str] | None = None,
) -> dict[str, Any]:
"""Resolve the deployment-configured transport without raising.
Returns the resolution rather than a bare string so a caller can tell an
unset value (which legitimately yields :data:`DEFAULT_TRANSPORT`) from a
configured value that is not permitted (which must fail closed, never
silently degrade to the default).
Args:
env: Environment mapping to read; defaults to ``os.environ``.
Returns:
dict with ``transport`` (normalized; the default when unset),
``configured``, ``source``, ``raw``, ``supported``, and ``reasons``.
"""
source_env = os.environ if env is None else env
raw = source_env.get(TRANSPORT_ENV)
normalized = normalize_transport(raw)
configured = bool(normalized)
if not configured:
return {
"transport": DEFAULT_TRANSPORT,
"configured": False,
"source": SOURCE_DEFAULT,
"raw": raw,
"supported": True,
"supported_transports": list(supported_transports()),
"env_key": TRANSPORT_ENV,
"reasons": [],
}
supported = normalized in SUPPORTED_TRANSPORTS
reasons: list[str] = []
if not supported:
reasons.append(
f"{TRANSPORT_ENV}={normalized!r} is not a registered MCP transport "
f"(#931). Registered: {list(supported_transports())}."
)
return {
"transport": normalized,
"configured": True,
"source": SOURCE_CONFIGURED,
"raw": raw,
"supported": supported,
"supported_transports": list(supported_transports()),
"env_key": TRANSPORT_ENV,
"reasons": reasons,
}
def require_configured_transport(env: Mapping[str, str] | None = None) -> str:
"""Resolved transport identifier, or raise when it is not permitted.
Raises:
TransportConfigurationError: the configured identifier is unregistered.
"""
resolution = resolve_configured_transport(env)
if not resolution["supported"]:
raise TransportConfigurationError("; ".join(resolution["reasons"]))
return str(resolution["transport"])
+924
View File
@@ -0,0 +1,924 @@
"""Transport-neutral MCP bind seam (#931).
These tests drive the *real* bind boundary — ``mark_sanctioned_daemon`` followed
by ``bind_native_mcp_transport`` from a canonical entrypoint path, with the
pytest allowance switched off — rather than mocking the new accessor. The
distinction matters here for the same reason it mattered in #941: a suite that
only exercises the helper in isolation cannot observe a seam that the live path
never reaches.
Covered:
1. no configured transport defaults to the local transport
2. explicit local transport binds
3. the sanctioned remote identifier binds through the seam
4. an unregistered identifier is rejected at bind time
5. an invalid bind prevents the server reaching tool service
6. the unbound state fails closed where a bind is required
7. every transport-aware guard reads the same authoritative value
8. client-controlled input cannot alter the bound transport
9. the durable decision-lock record carries the selected transport
10. existing stdio behaviour is unchanged
11. repeated / conflicting bind attempts follow one fail-closed contract
12. capability, role, repository and provenance protections do not regress
"""
from __future__ import annotations
import os
import tempfile
import unittest
from pathlib import Path
from unittest.mock import patch
REPO_ROOT = Path(__file__).resolve().parent.parent
import mcp_daemon_guard
import mcp_session_state
import mcp_transport_config
import irrecoverable_provenance
class _ProductionBind:
"""Context manager that reaches the real production bind path.
Patches only the two things a unit test cannot otherwise satisfy: the
resolved canonical entrypoint frame, and the pytest allowance that would
short-circuit ``mark_sanctioned_daemon``. Everything downstream of those —
validation, pinning, the rebind contract — runs unmodified.
"""
def __init__(self, env: dict[str, str] | None = None):
self._env = env or {}
self._stack: list = []
def __enter__(self):
mcp_daemon_guard.clear_native_runtime_for_tests()
canonical = str((REPO_ROOT / "mcp_server.py").resolve())
self._stack = [
patch.object(
mcp_daemon_guard,
"_caller_official_entrypoint_path",
side_effect=lambda: canonical,
),
patch.object(mcp_daemon_guard, "is_pytest_runtime", return_value=False),
patch.dict(os.environ, self._env),
]
for ctx in self._stack:
ctx.__enter__()
# Start from a clean configuration unless the test set one.
if mcp_transport_config.TRANSPORT_ENV not in self._env:
os.environ.pop(mcp_transport_config.TRANSPORT_ENV, None)
mcp_daemon_guard.mark_sanctioned_daemon()
return mcp_daemon_guard
def __exit__(self, *exc):
for ctx in reversed(self._stack):
ctx.__exit__(*exc)
mcp_daemon_guard.clear_native_runtime_for_tests()
return False
class TestPermittedSetIsSingleSourceOfTruth(unittest.TestCase):
"""AC3: identifiers are enumerated once, in the seam."""
def test_guard_allowlist_is_the_seam_allowlist(self):
self.assertIs(
mcp_daemon_guard._PRODUCTION_TRANSPORTS,
mcp_transport_config.SUPPORTED_TRANSPORTS,
)
def test_default_is_a_member_of_the_permitted_set(self):
self.assertIn(
mcp_transport_config.DEFAULT_TRANSPORT,
mcp_transport_config.SUPPORTED_TRANSPORTS,
)
def test_remote_identifier_is_permitted_and_not_the_default(self):
self.assertIn(
mcp_transport_config.REMOTE_TRANSPORT,
mcp_transport_config.SUPPORTED_TRANSPORTS,
)
self.assertNotEqual(
mcp_transport_config.REMOTE_TRANSPORT,
mcp_transport_config.DEFAULT_TRANSPORT,
)
self.assertTrue(
mcp_transport_config.is_remote_transport(
mcp_transport_config.REMOTE_TRANSPORT
)
)
def test_no_default_transport_literal_outside_the_seam(self):
"""AC3: no production module reads the literal outside the seam.
Review 635 flagged that a fixed five-module list cannot catch a *new*
module reintroducing the literal. This globs every production module in
the repository root instead, so the guarantee holds for code that does
not exist yet.
"""
default = mcp_transport_config.DEFAULT_TRANSPORT
needles = (f'"{default}"', f"'{default}'")
seam = Path(mcp_transport_config.__file__).name
scanned: list[str] = []
offenders: list[str] = []
for path in sorted(REPO_ROOT.glob("*.py")):
if path.name == seam:
continue # the seam is the one place the literal may live
scanned.append(path.name)
for lineno, line in enumerate(
path.read_text(encoding="utf-8").splitlines(), start=1
):
code = line.split("#", 1)[0]
if any(needle in code for needle in needles):
offenders.append(f"{path.name}:{lineno}: {line.strip()}")
# Guard the guard: a glob that silently matched nothing would pass.
self.assertGreater(len(scanned), 20, "production glob matched too little")
self.assertIn("mcp_daemon_guard.py", scanned)
self.assertIn("gitea_mcp_server.py", scanned)
self.assertEqual(offenders, [], "\n".join(offenders))
class TestConfiguredTransportResolution(unittest.TestCase):
"""AC1: configuration supplies the identifier; unset still yields the default."""
def test_unset_yields_default(self):
res = mcp_transport_config.resolve_configured_transport(env={})
self.assertEqual(res["transport"], mcp_transport_config.DEFAULT_TRANSPORT)
self.assertFalse(res["configured"])
self.assertEqual(res["source"], mcp_transport_config.SOURCE_DEFAULT)
self.assertTrue(res["supported"])
self.assertEqual(res["reasons"], [])
def test_blank_and_whitespace_are_treated_as_unset(self):
for raw in ("", " ", "\t\n"):
res = mcp_transport_config.resolve_configured_transport(
env={mcp_transport_config.TRANSPORT_ENV: raw}
)
self.assertEqual(res["transport"], mcp_transport_config.DEFAULT_TRANSPORT)
self.assertFalse(res["configured"])
def test_explicit_default_is_reported_as_configured(self):
res = mcp_transport_config.resolve_configured_transport(
env={
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.DEFAULT_TRANSPORT
)
}
)
self.assertEqual(res["transport"], mcp_transport_config.DEFAULT_TRANSPORT)
self.assertTrue(res["configured"])
self.assertEqual(res["source"], mcp_transport_config.SOURCE_CONFIGURED)
def test_remote_identifier_resolves_and_is_supported(self):
res = mcp_transport_config.resolve_configured_transport(
env={
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
)
self.assertEqual(res["transport"], mcp_transport_config.REMOTE_TRANSPORT)
self.assertTrue(res["supported"])
def test_case_and_padding_are_normalized(self):
padded = f" {mcp_transport_config.REMOTE_TRANSPORT.upper()} "
res = mcp_transport_config.resolve_configured_transport(
env={mcp_transport_config.TRANSPORT_ENV: padded}
)
self.assertEqual(res["transport"], mcp_transport_config.REMOTE_TRANSPORT)
self.assertTrue(res["supported"])
def test_unregistered_identifier_is_not_silently_defaulted(self):
res = mcp_transport_config.resolve_configured_transport(
env={mcp_transport_config.TRANSPORT_ENV: "carrier-pigeon"}
)
self.assertFalse(res["supported"])
self.assertEqual(res["transport"], "carrier-pigeon")
self.assertNotEqual(res["transport"], mcp_transport_config.DEFAULT_TRANSPORT)
self.assertTrue(res["reasons"])
def test_superseded_sse_transport_is_not_registered(self):
"""A real MCP transport that this deployment does not sanction."""
self.assertFalse(mcp_transport_config.is_supported_transport("sse"))
res = mcp_transport_config.resolve_configured_transport(
env={mcp_transport_config.TRANSPORT_ENV: "sse"}
)
self.assertFalse(res["supported"])
def test_require_configured_transport_raises_on_unregistered(self):
with self.assertRaises(mcp_transport_config.TransportConfigurationError):
mcp_transport_config.require_configured_transport(
env={mcp_transport_config.TRANSPORT_ENV: "carrier-pigeon"}
)
def test_non_string_configuration_never_matches_permitted_set(self):
for value in (object(), 1, None, True, ["stdio"], {"t": "stdio"}):
self.assertEqual(mcp_transport_config.normalize_transport(value), "")
self.assertFalse(mcp_transport_config.is_supported_transport(value))
class TestBindSeam(unittest.TestCase):
"""AC1/AC2: the live bind path resolves, validates, and pins."""
def tearDown(self) -> None:
mcp_daemon_guard.clear_native_runtime_for_tests()
def test_1_no_configured_transport_binds_default(self):
with _ProductionBind() as guard:
status = guard.bind_native_mcp_transport()
self.assertEqual(
status["transport"], mcp_transport_config.DEFAULT_TRANSPORT
)
self.assertEqual(
guard.bound_transport(), mcp_transport_config.DEFAULT_TRANSPORT
)
self.assertTrue(status["production_native_mcp_transport"])
def test_2_explicit_default_transport_binds(self):
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.DEFAULT_TRANSPORT
)
}
) as guard:
status = guard.bind_native_mcp_transport()
self.assertEqual(
status["transport"], mcp_transport_config.DEFAULT_TRANSPORT
)
self.assertTrue(guard.is_production_native_mcp_transport())
def test_2b_explicit_argument_still_binds(self):
"""The pre-#931 call form keeps working for launchers and tests."""
with _ProductionBind() as guard:
status = guard.bind_native_mcp_transport(
transport=mcp_transport_config.DEFAULT_TRANSPORT
)
self.assertEqual(
status["transport"], mcp_transport_config.DEFAULT_TRANSPORT
)
def test_3_sanctioned_remote_identifier_binds_through_the_seam(self):
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
status = guard.bind_native_mcp_transport()
self.assertEqual(
status["transport"], mcp_transport_config.REMOTE_TRANSPORT
)
self.assertEqual(
guard.bound_transport(), mcp_transport_config.REMOTE_TRANSPORT
)
# The remote identifier is trusted exactly like the local one; the
# listener that serves it is #938 and is not implemented here.
self.assertTrue(guard.is_native_mcp_transport())
self.assertTrue(guard.is_production_native_mcp_transport())
guard.assert_production_mutation_runtime("remote-bind")
def test_4_unregistered_identifier_rejected_at_bind_time(self):
with _ProductionBind(
{mcp_transport_config.TRANSPORT_ENV: "carrier-pigeon"}
) as guard:
with self.assertRaises(guard.UnsanctionedRuntimeError) as ctx:
guard.bind_native_mcp_transport()
self.assertIn("carrier-pigeon", str(ctx.exception))
self.assertIn("#931", str(ctx.exception))
# Nothing was bound, so nothing may dispatch.
self.assertIsNone(guard.bound_transport())
self.assertFalse(guard.is_native_mcp_transport())
def test_4b_unregistered_explicit_argument_rejected(self):
with _ProductionBind() as guard:
with self.assertRaises(guard.UnsanctionedRuntimeError):
guard.bind_native_mcp_transport(transport="carrier-pigeon")
self.assertIsNone(guard.bound_transport())
def test_4c_superseded_sse_rejected_at_bind_time(self):
with _ProductionBind({mcp_transport_config.TRANSPORT_ENV: "sse"}) as guard:
with self.assertRaises(guard.UnsanctionedRuntimeError):
guard.bind_native_mcp_transport()
self.assertIsNone(guard.bound_transport())
def test_5_invalid_bind_prevents_tool_service(self):
"""A failed bind must stop the server before it serves tools."""
with _ProductionBind(
{mcp_transport_config.TRANSPORT_ENV: "carrier-pigeon"}
) as guard:
with self.assertRaises(guard.UnsanctionedRuntimeError):
guard.bind_native_mcp_transport()
# This is the exact expression the entrypoint passes to mcp.run.
with self.assertRaises(guard.UnsanctionedRuntimeError) as ctx:
guard.assert_transport_bound("tool service")
self.assertIn("No MCP transport is bound", str(ctx.exception))
def test_6_unbound_state_fails_closed(self):
"""Entrypoint claimed but never bound — the offline-import shape."""
with _ProductionBind() as guard:
self.assertIsNone(guard.bound_transport())
self.assertFalse(guard.is_native_mcp_transport())
with self.assertRaises(guard.UnsanctionedRuntimeError):
guard.assert_transport_bound("tool service")
with self.assertRaises(guard.UnsanctionedRuntimeError):
guard.assert_sanctioned_mutation_runtime("gitea_mutation")
def test_6b_no_runtime_at_all_fails_closed(self):
mcp_daemon_guard.clear_native_runtime_for_tests()
with patch.object(mcp_daemon_guard, "is_pytest_runtime", return_value=False):
self.assertIsNone(mcp_daemon_guard.bound_transport())
with self.assertRaises(mcp_daemon_guard.UnsanctionedRuntimeError):
mcp_daemon_guard.assert_transport_bound("tool service")
class TestOneAuthoritativeValue(unittest.TestCase):
"""AC: every transport-aware guard observes the same value."""
def tearDown(self) -> None:
mcp_daemon_guard.clear_native_runtime_for_tests()
def test_7_all_guards_read_the_same_bound_value(self):
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
expected = mcp_transport_config.REMOTE_TRANSPORT
self.assertEqual(guard.bound_transport(), expected)
self.assertEqual(guard.assert_transport_bound(), expected)
self.assertEqual(guard.native_runtime_status()["bound_transport"], expected)
self.assertEqual(guard.native_runtime_status()["transport"], expected)
self.assertEqual(
guard.mutation_provenance_fields()["bound_transport"], expected
)
self.assertEqual(
irrecoverable_provenance.assess_transport_for_auth_mint()[
"bound_transport"
],
expected,
)
def test_8_environment_change_after_bind_cannot_move_the_value(self):
"""Client- or environment-shaped input must not alter a bound transport."""
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
self.assertEqual(
guard.bound_transport(), mcp_transport_config.DEFAULT_TRANSPORT
)
# A stray launcher (or an attacker) rewrites config post-bind.
os.environ[mcp_transport_config.TRANSPORT_ENV] = (
mcp_transport_config.REMOTE_TRANSPORT
)
self.assertEqual(
guard.bound_transport(), mcp_transport_config.DEFAULT_TRANSPORT
)
self.assertEqual(
guard.mutation_provenance_fields()["bound_transport"],
mcp_transport_config.DEFAULT_TRANSPORT,
)
os.environ[mcp_transport_config.TRANSPORT_ENV] = "carrier-pigeon"
self.assertEqual(
guard.bound_transport(), mcp_transport_config.DEFAULT_TRANSPORT
)
def test_8b_bound_transport_takes_no_caller_argument(self):
"""The accessor cannot be steered by a tool parameter."""
import inspect
self.assertEqual(
list(inspect.signature(mcp_daemon_guard.bound_transport).parameters), []
)
def test_11_rebinding_the_same_transport_is_idempotent(self):
with _ProductionBind() as guard:
first = guard.bind_native_mcp_transport()
second = guard.bind_native_mcp_transport()
self.assertEqual(first["transport"], second["transport"])
self.assertEqual(
guard.bound_transport(), mcp_transport_config.DEFAULT_TRANSPORT
)
def test_11b_rebinding_a_different_transport_fails_closed(self):
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
with self.assertRaises(guard.UnsanctionedRuntimeError) as ctx:
guard.bind_native_mcp_transport(
transport=mcp_transport_config.REMOTE_TRANSPORT
)
self.assertIn("already bound", str(ctx.exception))
# The first value survives the attempt.
self.assertEqual(
guard.bound_transport(), mcp_transport_config.DEFAULT_TRANSPORT
)
def test_11c_rebinding_an_unregistered_transport_fails_closed(self):
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
with self.assertRaises(guard.UnsanctionedRuntimeError):
guard.bind_native_mcp_transport(transport="carrier-pigeon")
self.assertEqual(
guard.bound_transport(), mcp_transport_config.DEFAULT_TRANSPORT
)
class TestDurableRecordCarriesTransport(unittest.TestCase):
"""AC4: the identifier reaches a durable decision-lock record."""
def tearDown(self) -> None:
mcp_daemon_guard.clear_native_runtime_for_tests()
def _save_and_read_decision_lock(self, state_dir: str) -> dict:
mcp_session_state.save_state(
kind=mcp_session_state.KIND_DECISION_LOCK,
payload={"pr_number": 931, "action": "COMMENT"},
remote="prgs",
org="Scaled-Tech-Consulting",
repo="Gitea-Tools",
profile_identity="prgs-author",
state_dir=state_dir,
)
loaded = mcp_session_state.load_state(
kind=mcp_session_state.KIND_DECISION_LOCK,
remote="prgs",
org="Scaled-Tech-Consulting",
repo="Gitea-Tools",
profile_identity="prgs-author",
state_dir=state_dir,
)
self.assertIsNotNone(loaded)
return loaded
def test_9_decision_lock_records_the_bound_transport(self):
with tempfile.TemporaryDirectory() as tmp:
mcp_daemon_guard.clear_native_runtime_for_tests()
mcp_daemon_guard.install_test_native_runtime()
record = self._save_and_read_decision_lock(tmp)
self.assertIn("bound_transport", record)
self.assertEqual(
record["bound_transport"], mcp_daemon_guard.bound_transport()
)
def test_9b_unbound_runtime_records_no_transport_identifier(self):
with tempfile.TemporaryDirectory() as tmp:
mcp_daemon_guard.clear_native_runtime_for_tests()
record = self._save_and_read_decision_lock(tmp)
self.assertIn("bound_transport", record)
self.assertIsNone(record["bound_transport"])
def test_9c_provenance_fields_expose_the_identifier(self):
mcp_daemon_guard.clear_native_runtime_for_tests()
fields = mcp_daemon_guard.mutation_provenance_fields()
self.assertIn("bound_transport", fields)
self.assertIsNone(fields["bound_transport"])
class TestStdioBehaviourUnchanged(unittest.TestCase):
"""AC6 / prompt items 10 and 12: no regression on the existing path."""
def tearDown(self) -> None:
mcp_daemon_guard.clear_native_runtime_for_tests()
def test_10_trust_class_field_keeps_its_pre_931_values(self):
"""``transport`` remains the trust class, not the identifier."""
mcp_daemon_guard.clear_native_runtime_for_tests()
self.assertEqual(
mcp_daemon_guard.mutation_provenance_fields()["transport"], "untrusted"
)
mcp_daemon_guard.install_test_native_runtime()
self.assertEqual(
mcp_daemon_guard.mutation_provenance_fields()["transport"],
"test_native_mcp",
)
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
self.assertEqual(
guard.mutation_provenance_fields()["transport"], "native_mcp"
)
def test_10b_default_bind_reproduces_the_pre_931_runtime_record(self):
with _ProductionBind() as guard:
status = guard.bind_native_mcp_transport()
self.assertTrue(status["native_mcp_transport"])
self.assertTrue(status["production_native_mcp_transport"])
self.assertEqual(status["mode"], "production")
self.assertEqual(status["phase"], "transport_bound")
self.assertEqual(
status["transport"], mcp_transport_config.DEFAULT_TRANSPORT
)
guard.assert_sanctioned_mutation_runtime("native-ide")
guard.assert_production_mutation_runtime("native-ide")
def test_12_session_state_root_still_pinned_at_bind(self):
"""#695 AC2 must survive the seam."""
with tempfile.TemporaryDirectory() as legit:
with tempfile.TemporaryDirectory() as rogue:
with _ProductionBind(
{mcp_daemon_guard.SESSION_STATE_DIR_ENV: legit}
) as guard:
guard.bind_native_mcp_transport()
self.assertEqual(
guard.pinned_session_state_dir(), str(Path(legit).resolve())
)
os.environ[mcp_daemon_guard.SESSION_STATE_DIR_ENV] = rogue
self.assertEqual(
guard.pinned_session_state_dir(), str(Path(legit).resolve())
)
def test_12b_test_mode_record_still_cannot_authorize_production(self):
mcp_daemon_guard.clear_native_runtime_for_tests()
mcp_daemon_guard.install_test_native_runtime()
self.assertTrue(mcp_daemon_guard.is_native_mcp_transport())
self.assertFalse(mcp_daemon_guard.is_production_native_mcp_transport())
with self.assertRaises(mcp_daemon_guard.UnsanctionedRuntimeError):
mcp_daemon_guard.assert_production_mutation_runtime("prod-endpoint")
def test_12c_bind_still_requires_the_canonical_entrypoint(self):
"""A remote identifier does not relax provenance."""
mcp_daemon_guard.clear_native_runtime_for_tests()
with patch.object(mcp_daemon_guard, "is_pytest_runtime", return_value=False):
with patch.object(
mcp_daemon_guard,
"_caller_official_entrypoint_path",
return_value=None,
):
with self.assertRaises(mcp_daemon_guard.UnsanctionedRuntimeError):
mcp_daemon_guard.bind_native_mcp_transport(
transport=mcp_transport_config.REMOTE_TRANSPORT
)
self.assertIsNone(mcp_daemon_guard.bound_transport())
def test_12d_bind_still_requires_a_prior_entrypoint_claim(self):
mcp_daemon_guard.clear_native_runtime_for_tests()
canonical = str((REPO_ROOT / "mcp_server.py").resolve())
with patch.object(mcp_daemon_guard, "is_pytest_runtime", return_value=False):
with patch.object(
mcp_daemon_guard,
"_caller_official_entrypoint_path",
side_effect=lambda: canonical,
):
# No mark_sanctioned_daemon() first.
with self.assertRaises(
mcp_daemon_guard.UnsanctionedRuntimeError
) as ctx:
mcp_daemon_guard.bind_native_mcp_transport()
self.assertIn("no entrypoint claim", str(ctx.exception))
def test_12e_auth_mint_verdict_is_unchanged_for_the_default_transport(self):
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
verdict = irrecoverable_provenance.assess_transport_for_auth_mint()
self.assertTrue(verdict["allowed"])
self.assertTrue(verdict["native_mcp_transport"])
self.assertTrue(verdict["production_native_mcp_transport"])
self.assertEqual(verdict["reasons"], [])
def test_12f_auth_mint_still_refuses_an_unbound_runtime(self):
with _ProductionBind():
# Claimed but never bound.
verdict = irrecoverable_provenance.assess_transport_for_auth_mint()
self.assertFalse(verdict["allowed"])
self.assertTrue(verdict["reasons"])
self.assertIsNone(verdict["bound_transport"])
class TestExecutionBoundary(unittest.TestCase):
"""Recognition is not execution authorization (#931, review 635 B1/B2).
A registered remote identifier must still bind, pin and record — the #931
seam — while being refused at the serve boundary, because serving it would
start a listener with no authentication or per-request principal. That
listener belongs to #938.
"""
def tearDown(self) -> None:
mcp_daemon_guard.clear_native_runtime_for_tests()
# -- the two sets are distinct, and narrower in the right direction ----
def test_executable_set_is_a_strict_subset_of_recognized(self):
self.assertTrue(
mcp_transport_config.EXECUTABLE_TRANSPORTS
< mcp_transport_config.SUPPORTED_TRANSPORTS
)
def test_default_transport_is_executable(self):
self.assertTrue(
mcp_transport_config.is_executable_transport(
mcp_transport_config.DEFAULT_TRANSPORT
)
)
def test_remote_transport_is_recognized_but_not_executable(self):
remote = mcp_transport_config.REMOTE_TRANSPORT
self.assertTrue(mcp_transport_config.is_supported_transport(remote))
self.assertFalse(mcp_transport_config.is_executable_transport(remote))
def test_remote_listener_ownership_is_declared(self):
self.assertEqual(
mcp_transport_config.TRANSPORT_EXECUTION_OWNER[
mcp_transport_config.REMOTE_TRANSPORT
],
"#938",
)
# -- stdio still reaches the runner, unchanged -------------------------
def test_default_transport_is_authorized_for_service(self):
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
self.assertEqual(
guard.authorize_transport_execution("tool service"),
mcp_transport_config.DEFAULT_TRANSPORT,
)
self.assertTrue(guard.assess_serve_authorization()["allowed"])
def test_default_transport_reaches_the_real_runner(self):
"""The production runner is actually invoked, with stdio, unchanged."""
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
seen = {}
class _Runner:
def run(self, transport=None, **kw):
seen["transport"] = transport
_Runner().run(transport=guard.authorize_transport_execution("tool service"))
self.assertEqual(
seen["transport"], mcp_transport_config.DEFAULT_TRANSPORT
)
# -- streamable-http binds, records, and is refused before serving -----
def test_remote_transport_binds_and_is_recorded(self):
"""The #931 seam is intact: it binds, pins and records."""
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
self.assertEqual(
guard.bound_transport(), mcp_transport_config.REMOTE_TRANSPORT
)
self.assertEqual(
guard.mutation_provenance_fields()["bound_transport"],
mcp_transport_config.REMOTE_TRANSPORT,
)
def test_remote_transport_is_refused_at_the_serve_boundary(self):
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
with self.assertRaises(guard.TransportExecutionError) as ctx:
guard.authorize_transport_execution("tool service")
err = ctx.exception
self.assertEqual(
err.blocker_kind,
mcp_transport_config.BLOCKER_LISTENER_NOT_COMMISSIONED,
)
self.assertEqual(err.owner_issue, "#938")
self.assertEqual(err.transport, mcp_transport_config.REMOTE_TRANSPORT)
def test_refusal_names_transport_and_unmet_requirement_and_owner(self):
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
with self.assertRaises(guard.TransportExecutionError) as ctx:
guard.authorize_transport_execution("tool service")
text = str(ctx.exception)
self.assertIn(mcp_transport_config.REMOTE_TRANSPORT, text)
self.assertIn("#938", text)
self.assertIn("not commissioned", text)
self.assertIn("#931", text)
def test_remote_transport_never_reaches_the_runner(self):
"""No transport value is handed to a run() call for the remote case."""
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
calls = []
class _Runner:
def run(self, transport=None, **kw):
calls.append(transport)
with self.assertRaises(guard.TransportExecutionError):
_Runner().run(
transport=guard.authorize_transport_execution("tool service")
)
self.assertEqual(calls, [], "runner must never be invoked")
def test_no_http_listener_is_created_for_remote_transport(self):
"""Nothing in the refusal path touches uvicorn or a socket bind."""
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
import socket
bound_sockets = []
real_bind = socket.socket.bind
def _tripwire(self, addr): # pragma: no cover - must not run
bound_sockets.append(addr)
return real_bind(self, addr)
with patch.object(socket.socket, "bind", _tripwire):
with self.assertRaises(guard.TransportExecutionError):
guard.authorize_transport_execution("tool service")
self.assertEqual(bound_sockets, [], "no socket may be bound")
def test_no_mutation_is_authorized_after_the_denial(self):
"""The denial leaves no partial state that would let a tool dispatch."""
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
with self.assertRaises(guard.TransportExecutionError):
guard.authorize_transport_execution("tool service")
# Serve stays unauthorized on every subsequent query.
self.assertFalse(guard.assess_serve_authorization()["allowed"])
self.assertFalse(guard.native_runtime_status()["serve_authorized"])
with self.assertRaises(guard.TransportExecutionError):
guard.authorize_transport_execution("tool service")
# -- B2: the decision genuinely consumes bound_transport ---------------
def test_serve_decision_consumes_bound_transport(self):
"""Changing only bound_transport flips the verdict."""
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
self.assertTrue(guard.assess_serve_authorization()["allowed"])
with patch.object(
guard,
"bound_transport",
return_value=mcp_transport_config.REMOTE_TRANSPORT,
):
verdict = guard.assess_serve_authorization()
self.assertFalse(verdict["allowed"])
self.assertEqual(
verdict["transport"], mcp_transport_config.REMOTE_TRANSPORT
)
with self.assertRaises(guard.TransportExecutionError):
guard.authorize_transport_execution("tool service")
def test_serve_verdict_reports_the_bound_transport(self):
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
verdict = guard.assess_serve_authorization()
self.assertEqual(
verdict["transport"], mcp_transport_config.REMOTE_TRANSPORT
)
self.assertTrue(verdict["recognized"])
self.assertFalse(verdict["executable"])
# -- earlier and later boundaries are unchanged ------------------------
def test_unregistered_identifier_still_fails_at_bind_not_at_serve(self):
with _ProductionBind(
{mcp_transport_config.TRANSPORT_ENV: "carrier-pigeon"}
) as guard:
with self.assertRaises(guard.UnsanctionedRuntimeError) as ctx:
guard.bind_native_mcp_transport()
self.assertIn("not a registered MCP transport", str(ctx.exception))
self.assertIsNone(guard.bound_transport())
def test_unbound_execution_keeps_the_pre_existing_failure(self):
with _ProductionBind() as guard:
with self.assertRaises(guard.UnsanctionedRuntimeError) as ctx:
guard.authorize_transport_execution("tool service")
self.assertIn("No MCP transport is bound", str(ctx.exception))
self.assertNotIsInstance(ctx.exception, guard.TransportExecutionError)
def test_transport_execution_error_is_caught_by_existing_handlers(self):
"""Subclassing keeps every pre-existing fail-closed handler correct."""
self.assertTrue(
issubclass(
mcp_daemon_guard.TransportExecutionError,
mcp_daemon_guard.UnsanctionedRuntimeError,
)
)
def test_test_mode_runtime_is_not_servable(self):
"""The pytest-only record is outside the recognized and executable sets."""
mcp_daemon_guard.clear_native_runtime_for_tests()
mcp_daemon_guard.install_test_native_runtime()
self.assertNotIn(
mcp_daemon_guard.bound_transport(),
mcp_transport_config.SUPPORTED_TRANSPORTS,
)
self.assertFalse(mcp_daemon_guard.assess_serve_authorization()["allowed"])
def test_serve_authorization_does_not_leak_across_runtimes(self):
"""A later runtime's verdict never reflects an earlier one."""
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
)
}
) as guard:
guard.bind_native_mcp_transport()
self.assertFalse(guard.assess_serve_authorization()["allowed"])
with _ProductionBind() as guard:
guard.bind_native_mcp_transport()
self.assertTrue(guard.assess_serve_authorization()["allowed"])
mcp_daemon_guard.clear_native_runtime_for_tests()
self.assertEqual(
mcp_daemon_guard.assess_serve_authorization()["blocker_kind"],
mcp_transport_config.BLOCKER_TRANSPORT_NOT_BOUND,
)
def test_refusal_carries_no_credential_material(self):
with _ProductionBind(
{
mcp_transport_config.TRANSPORT_ENV: (
mcp_transport_config.REMOTE_TRANSPORT
),
"GITEA_TOKEN": "super-secret-value",
}
) as guard:
guard.bind_native_mcp_transport()
with self.assertRaises(guard.TransportExecutionError) as ctx:
guard.authorize_transport_execution("tool service")
blob = str(ctx.exception) + repr(ctx.exception.assessment)
self.assertNotIn("super-secret-value", blob)
self.assertNotIn("GITEA_TOKEN", blob)
class TestEntrypointWiring(unittest.TestCase):
"""The live entrypoint must use the seam and the execution guard."""
def test_entrypoint_binds_without_a_literal_transport(self):
text = (REPO_ROOT / "gitea_mcp_server.py").read_text(encoding="utf-8")
self.assertIn("mcp_daemon_guard.bind_native_mcp_transport()", text)
self.assertNotIn('bind_native_mcp_transport(transport="stdio")', text)
def test_entrypoint_serves_only_through_the_execution_guard(self):
text = (REPO_ROOT / "gitea_mcp_server.py").read_text(encoding="utf-8")
self.assertIn(
"mcp.run(transport=mcp_daemon_guard.authorize_transport_execution", text
)
self.assertNotIn('mcp.run(transport="stdio")', text)
self.assertNotIn(
"mcp.run(transport=mcp_daemon_guard.assert_transport_bound", text
)
def test_every_run_call_in_production_goes_through_the_guard(self):
"""Glob the production surface: no serve site may bypass the guard."""
offenders: list[str] = []
run_sites = 0
for path in sorted(REPO_ROOT.glob("*.py")):
for lineno, line in enumerate(
path.read_text(encoding="utf-8").splitlines(), start=1
):
code = line.split("#", 1)[0]
if "mcp.run(" not in code:
continue
run_sites += 1
if "authorize_transport_execution" not in code:
offenders.append(f"{path.name}:{lineno}: {line.strip()}")
self.assertEqual(run_sites, 1, "expected exactly one serve site")
self.assertEqual(offenders, [], "\n".join(offenders))
if __name__ == "__main__":
unittest.main()
+323
View File
@@ -0,0 +1,323 @@
"""Validation tooling for the remote-MCP threat model (#956).
#956 requires that "every boundary claim [is] traceable to a file and line
anchor that resolves at the reviewed commit". A prose document cannot enforce
that about itself, and #930 demonstrated the failure mode: its inventory cited
``gitea_mcp_server.py`` anchors generated at ``7bf4f125`` which no longer point
at the described code at ``aad5c8b4``. Nothing failed, because nothing checked.
These tests are that check. They enforce, in both directions:
* every ``file.py:NNN`` anchor cited in the prose is declared in the fixture;
* every declared anchor resolves — the file exists, the line exists, and the
source line actually contains the substring the fixture claims for it;
* the document's structural obligations (assets, adversaries, boundaries,
credential rows, the co-residency ruling, and the child mapping) are present
and internally consistent.
A refactor that shifts a line number therefore breaks the suite instead of
silently rotting the security documentation.
"""
import json
import os
import re
import unittest
REPO_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DOC_PATH = os.path.join(REPO_ROOT, "docs", "remote-mcp", "threat-model.md")
FIXTURE_PATH = os.path.join(
REPO_ROOT, "docs", "remote-mcp", "threat-model-anchors.json"
)
# ``module.py:123`` as it appears inside markdown inline code spans.
ANCHOR_RE = re.compile(r"`([A-Za-z0-9_./-]+\.py):(\d+)`")
# The epic children this document must map to a boundary (#929 children 2-10).
REQUIRED_CHILDREN = [931, 932, 933, 934, 935, 936, 937, 938, 939]
# The adversaries #956 names explicitly.
REQUIRED_ADVERSARIES = [
"compromised LLM client",
"prompt injection",
"malicious tool arguments",
"network attacker",
"curious operator",
]
def _read(path):
with open(path, "r", encoding="utf-8") as fh:
return fh.read()
def _heading_re(title):
"""Match a level-2 heading by title, with or without section numbering.
The document numbers its sections ('## 6. Decomposition ruling'), so an
exact-substring assertion would break on renumbering without the document
having actually lost anything.
"""
return re.compile(
r"^##\s+(?:\d+\.\s+)?" + re.escape(title), re.MULTILINE
)
def _section_body(doc, title):
"""Return the text of section *title*, bounded by the next level-2 heading.
Bounding matters: an unbounded slice runs to end-of-document, so the
walkthrough tables in a later section leak into the child-to-boundary
mapping and satisfy its coverage check with rows that assign no owner.
"""
match = _heading_re(title).search(doc)
if match is None:
return None
rest = doc[match.end():]
nxt = re.search(r"^##\s", rest, re.MULTILINE)
return rest[: nxt.start()] if nxt else rest
def _source_line(rel_path, lineno):
"""Return the 1-based *lineno* of *rel_path*, or None if out of range."""
abs_path = os.path.join(REPO_ROOT, rel_path)
if not os.path.exists(abs_path):
return None
with open(abs_path, "r", encoding="utf-8", errors="replace") as fh:
for idx, line in enumerate(fh, start=1):
if idx == lineno:
return line
return None
class ThreatModelFixtureTests(unittest.TestCase):
"""The fixture itself must be well-formed before it can prove anything."""
def setUp(self):
self.fixture = json.loads(_read(FIXTURE_PATH))
def test_fixture_declares_a_generation_commit(self):
sha = self.fixture.get("generated_against_commit") or ""
self.assertRegex(
sha,
r"^[0-9a-f]{40}$",
"the fixture must record the full commit its anchors were taken at",
)
def test_fixture_anchors_are_unique_and_well_formed(self):
seen = set()
for entry in self.fixture["anchors"]:
anchor = entry["anchor"]
self.assertNotIn(anchor, seen, f"duplicate anchor entry: {anchor}")
seen.add(anchor)
self.assertRegex(anchor, r"^[A-Za-z0-9_./-]+\.py:[1-9]\d*$", anchor)
self.assertTrue(
(entry.get("expect") or "").strip(),
f"anchor {anchor} declares no 'expect' substring, so it proves nothing",
)
class ThreatModelAnchorResolutionTests(unittest.TestCase):
"""#956 required positive test: every anchor resolves at the reviewed commit."""
def setUp(self):
self.fixture = json.loads(_read(FIXTURE_PATH))
self.doc = _read(DOC_PATH)
def test_every_declared_anchor_resolves_to_the_claimed_source_line(self):
failures = []
for entry in self.fixture["anchors"]:
rel_path, _, raw_lineno = entry["anchor"].partition(":")
lineno = int(raw_lineno)
line = _source_line(rel_path, lineno)
if line is None:
failures.append(f"{entry['anchor']}: file or line does not exist")
continue
if entry["expect"] not in line:
failures.append(
f"{entry['anchor']}: expected {entry['expect']!r}, "
f"found {line.strip()!r}"
)
self.assertEqual(
[], failures, "unresolved threat-model anchors:\n" + "\n".join(failures)
)
def test_every_anchor_cited_in_the_document_is_declared_in_the_fixture(self):
declared = {e["anchor"] for e in self.fixture["anchors"]}
cited = {f"{m.group(1)}:{m.group(2)}" for m in ANCHOR_RE.finditer(self.doc)}
undeclared = sorted(cited - declared)
self.assertEqual(
[],
undeclared,
"document cites anchors that no test verifies: " + ", ".join(undeclared),
)
def test_the_document_actually_cites_anchors(self):
cited = {f"{m.group(1)}:{m.group(2)}" for m in ANCHOR_RE.finditer(self.doc)}
self.assertGreaterEqual(
len(cited),
30,
"a boundary document with almost no anchors is not traceable",
)
def test_unresolvable_anchor_is_detected(self):
"""Negative control: the checker must fail on a deliberately bad anchor.
Without this, a checker that silently passed everything would look
identical to a correct one.
"""
self.assertIsNone(_source_line("gitea_config.py", 10**9))
self.assertIsNone(_source_line("no_such_module_for_956.py", 1))
real = _source_line("gitea_config.py", 54)
self.assertIsNotNone(real)
self.assertNotIn("this substring is not on that line", real)
class ThreatModelStructureTests(unittest.TestCase):
"""The document must contain what #956's acceptance criteria demand."""
def setUp(self):
self.doc = _read(DOC_PATH)
def test_records_the_commit_it_was_generated_against(self):
fixture = json.loads(_read(FIXTURE_PATH))
self.assertIn(
fixture["generated_against_commit"],
self.doc,
"the document must state the commit its anchors resolve at",
)
def test_names_every_required_adversary(self):
low = self.doc.lower()
for adversary in REQUIRED_ADVERSARIES:
self.assertIn(adversary.lower(), low, f"adversary not covered: {adversary}")
def test_maps_every_epic_child_from_two_through_ten(self):
for number in REQUIRED_CHILDREN:
self.assertIn(
f"#{number}",
self.doc,
f"epic child #{number} is not mapped to a boundary",
)
def test_credential_rows_declare_holder_boundary_and_blast_radius(self):
for column in ("Holder", "Boundary", "Blast radius"):
self.assertIn(
column,
self.doc,
f"the credential inventory must state each credential's {column.lower()}",
)
def test_states_an_explicit_co_residency_ruling(self):
"""AC3/AC5: an explicit ruling, not an implication."""
self.assertIsNotNone(
_heading_re("Decomposition ruling").search(self.doc),
"the document must contain an explicit decomposition-ruling section",
)
for service in ("Jenkins", "GlitchTip", "Sentry", "database"):
self.assertIn(service, self.doc, f"ruling does not address {service}")
self.assertRegex(
self.doc,
r"D1\b.*must not",
"the ruling must state the prohibition, not merely discuss it",
)
def test_contains_the_compromised_client_walkthrough(self):
"""#956 required negative/adversarial test."""
self.assertIsNotNone(
_heading_re("Adversarial walkthrough").search(self.doc),
"the required compromised-client walkthrough is missing",
)
self.assertIn("Before the migration", self.doc)
self.assertIn("After the migration", self.doc)
def test_every_boundary_states_what_it_protects_and_what_crossing_requires(self):
boundary_ids = set(re.findall(r"\bB(\d+)\b", self.doc))
self.assertGreaterEqual(
len(boundary_ids), 5, "too few trust boundaries to be a decomposition"
)
for column in (
"Protects",
"Crossing requires today",
"Crossing must require remotely",
):
self.assertIn(column, self.doc, f"boundary table is missing '{column}'")
def test_declares_itself_documentation_only(self):
self.assertIn("documentation only", self.doc.lower())
class ThreatModelConsistencyTests(unittest.TestCase):
"""Counts stated in prose must match the rows actually present."""
def setUp(self):
self.doc = _read(DOC_PATH)
def _declared_ids(self, prefix):
# Table rows begin '| CR1 |' / '| B3 |' / '| A2 |'.
return sorted(
{
int(m)
for m in re.findall(
r"^\|\s*%s(\d+)\s*\|" % prefix, self.doc, re.MULTILINE
)
}
)
def test_identifier_sequences_have_no_gaps(self):
for prefix, label in (
("A", "assets"),
("B", "boundaries"),
("CR", "credentials"),
):
ids = self._declared_ids(prefix)
self.assertTrue(ids, f"no {label} declared")
self.assertEqual(
list(range(1, len(ids) + 1)),
ids,
f"{label} identifiers must run 1..n with no gaps; got {ids}",
)
def test_stated_credential_count_matches_the_rows(self):
ids = self._declared_ids("CR")
match = re.search(r"(\d+)\s+credential(?:s)? in total", self.doc)
self.assertIsNotNone(match, "the credential inventory must state its own total")
self.assertEqual(
len(ids),
int(match.group(1)),
"stated credential total disagrees with the number of rows",
)
def test_every_boundary_is_owned_by_at_least_one_child(self):
"""Each boundary must be owned by a child *in the mapping table*.
Scanning the whole section would let a prose summary line ("Boundary
coverage: ... B5 (#936)") satisfy the assertion while the table row
that actually assigns the owner had been emptied — verified by
deliberately blanking a row and watching a whole-section check still
pass. Only table rows count.
"""
mapping_section = _section_body(self.doc, "Child-to-boundary mapping")
self.assertIsNotNone(
mapping_section, "child-to-boundary mapping section is missing"
)
rows = [
line
for line in mapping_section.splitlines()
if line.lstrip().startswith("|") and re.search(r"#93\d", line)
]
self.assertGreaterEqual(
len(rows), len(REQUIRED_CHILDREN), "mapping table has too few child rows"
)
mapped = set(re.findall(r"\bB(\d+)\b", "\n".join(rows)))
declared = {str(i) for i in self._declared_ids("B")}
unmapped = sorted(declared - mapped, key=int)
self.assertEqual(
[],
unmapped,
"boundaries with no owning child: " + ", ".join("B" + u for u in unmapped),
)
if __name__ == "__main__":
unittest.main()