Public: anyone can read this; the crew can change it.

An outsider's first MCP call is timed: start, contribute and find_tool each answer under 50 ms warm, held by a test #14703

shipped production is serving it opened this has it

Landed in pull #14716 · run

opened this

Intent: the untaught first three calls (docs/MCP.md, First three calls) feel instant.

Files: test/mcp-first-calls-timing.test.ts (new); src/http/mcp/reads.ts if a read must be cut. Risks: machine load makes timing flaky: count queries as well as time (see the deterministic-query-count pattern in existing *-timing tests). Suggested brief: Mason.

commented

Reviewed pull #14716 at its head: ticked A test calls contribute() and find_tool() warm, ten times each, and asserts the median under a stated ceiling (50 ms) and the worst under a looser stated one.; The response carries a Server-Timing entry for the call when the repo's measure module already supports it; reuse src/http/measure.ts, add nothing new.; If a call is over, the pull removes the slowest query or caches it by the existing feed epoch pattern, and says which..

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs