fix: keep read-only MCP tools responsive#131
Conversation
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8c65eaeebd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
5217b6e to
bcf5738
Compare
8c65eae to
5af922f
Compare
bcf5738 to
9f65011
Compare
5af922f to
1c30d1f
Compare
9f65011 to
9ad434e
Compare
8f67de4 to
43fe723
Compare
a18f595 to
5032fe1
Compare
43fe723 to
07fbdd4
Compare
5032fe1 to
7eb7678
Compare
07fbdd4 to
a400b84
Compare
00aba68 to
d9e7dd1
Compare
This PR stacks on #130. Review commit/diff against
split/self-improving-04-dashboard-ui.Summary
node_modules.Root cause
Read-only MCP calls could create/open a project server and trigger a startup or pre-dispatch staleness sync. On huge indexes this meant simple calls like
tracedecay_storage_statusortracedecay_searchcould spend minutes walking the filesystem and holding the sync lock before returning any result.Verification
cargo test --lib read_only_tools_skip_lazy_sync_before_dispatchcargo test --test mcp_handler_test wait_for_startup_catch_up_waits_for_transcript_ingest_flagcargo test --test multi_mcp_coordination_testgit diff --check