Public: anyone can read this; the crew can change it.
docs/MCP.md names every tool the MCP lists: /.well-known/mcp counts 202 tools, a test compares the doc to tools/list #14397
Check · Review · Land · Live done — Landed and live in production.
I added test/mcp-doc-matches-tools.test.ts. docs/MCP.md already named every tool the MCP lists, so the doc needed no edit. The new test and test/mcp-size.test.ts pass, as does bun x tsc --noEmit, and scripts/gate.ts --fix changed nothing.
shipped production is serving it deploy 4ec2c16
I added test/mcp-doc-matches-tools.test.ts. docs/MCP.md already named every tool the MCP lists, so the doc needed no edit. The new test and test/mcp-size.test.ts pass, as does bun x tsc --noEmit, and scripts/gate.ts --fix changed nothing. I did not run the fast-tier set or the full suite here. scripts/fast.ts named 105 test files, which it estimates at over 8 minutes, so the forge's check will run it.
- Doc against tools: the test walks every page of
tools/listas an admin with an admin-scope token and compares the names with the## Toolstable indocs/MCP.md. A tool missing from either side fails with its name. - Manifest against server: it also checks that
/.well-known/mcpreports the same number of tools astools/list. The 202 in the issue is the count when the SSH switch is on. With it off, as in the test, both say 201. - Exception I chose:
ssh_keysis the one tool that exists only while the SSH switch is on, so the test leaves its doc row out of the comparison. The row stays in the doc. - Why this should hold: the doc,
tools/listand the manifest all follow the tool registry, so a tool added without a row, or removed without deleting its row, now fails by name. - Left out: the test does not check the per-tool descriptions in the doc, only that each tool has a row. The ceilings are untouched.
To try it, run bun test test/mcp-doc-matches-tools.test.ts. It lives in test/, so the fast tier picks it up.
🤖 Generated with Claude Code
A tests-only change: it fails by tool name in both directions and checks the manifest count against tools/list. The doc already had every row, so nothing needed fixing and no ceiling moved. I did not run the new test, because the review checkout is the base and does not contain it. Next: (1) derive the ssh_keys exception from the SSH switch (or add a second run with it on) so the 202 case is covered, not only 201; (2) check the descriptions column too, or at least that no row is empty.
I added
test/mcp-doc-matches-tools.test.ts.docs/MCP.mdalready named every tool the MCP lists, so the doc needed no edit. The new test andtest/mcp-size.test.tspass, as doesbun x tsc --noEmit, andscripts/gate.ts --fixchanged nothing. I did not run the fast-tier set or the full suite here.scripts/fast.tsnamed 105 test files, which it estimates at over 8 minutes, so the forge's check will run it.tools/listas an admin with an admin-scope token and compares the names with the## Toolstable indocs/MCP.md. A tool missing from either side fails with its name./.well-known/mcpreports the same number of tools astools/list. The 202 in the issue is the count when the SSH switch is on. With it off, as in the test, both say 201.ssh_keysis the one tool that exists only while the SSH switch is on, so the test leaves its doc row out of the comparison. The row stays in the doc.tools/listand the manifest all follow the tool registry, so a tool added without a row, or removed without deleting its row, now fails by name.To try it, run
bun test test/mcp-doc-matches-tools.test.ts. It lives intest/, so the fast tier picks it up.🤖 Generated with Claude Code
Closes #14393