stacklok / toolhive / 25675148336

66%

Build Type

push

github

Committed by web-flow

Commit Message

Tolerate spec-violating list methods on backend init (#5232)

* Tolerate spec-violating list methods on backend init

When a backend advertises resources or prompts capability in its
initialize response but returns JSON-RPC -32601 to resources/list or
prompts/list, treat that as "the backend has no resources/prompts"
rather than a fatal init error. This unblocks third-party MCP servers
(e.g. Atlassian Rovo) whose initialize response contradicts their
implemented method set, so users still get the backend's tools instead
of silently losing them.

Implements changes for issue #5231:
- Recover from errors.Is(err, mcp.ErrMethodNotFound) on resources/list
  and prompts/list with a WARN log naming the backend and method.
- Keep tools/list failures and non-(-32601) errors from list methods
  fatal so we are not silencing arbitrary failures.
- Add table-driven unit tests with a fake JSON-RPC backend covering
  the recoverable, fatal, and regression-guard cases.

* Address PR #5232 review feedback

Addresses stacklok/toolhive#5232 review comments:
- MEDIUM mcp_session_capabilities_test.go (3211850392): add success-path
  rows for resources and prompts so the switch's default arm (the populate
  loop) is exercised, including BackendID field-mapping assertions.
- LOW mcp_session.go (3211850393): tighten the recovery comment in both
  the resources and prompts switch arms to a single rationale-focused
  sentence.
- LOW mcp_session_capabilities_test.go (3211850399): replace
  wantToolsCalled bool with wantToolsCalls int so the assertion shape
  matches the resources/prompts counters and zero-call cases become
  expressible.
- LOW mcp_session_capabilities_test.go (3211850401): use
  assert.ErrorContains in place of assert.True(strings.Contains(...))
  and drop the now-unused strings import.
- LOW mcp_session.go (3211850404): enrich the WARN log in both arms
  with workload name and base URL so the breadcrumb is grep-friendly
  without an ID-to-name lookup.
- LOW mc... (continued)

Coverage Stats

51 of 51 new or added lines in 1 file covered. (100.0%)

2 existing lines in 1 file now uncovered.

64024 of 98519 relevant lines covered (64.99%)

61.61 hits per line