MCP `apps` (or a `pulls` action) reads an adapter's recorded pull runs, as the API does #10119

MCP apps now reads an adapter's recorded pull runs: call it with { repo: "owner/name", pulls: true, limit? }. The new test, the type check, test/mcp-size.test.ts and test/mcp-apps-archived.test.ts pass. I ran bun scripts/gate.ts but didn't read its output, so treat that as unchecked. My first push left a type error in the test; the second push fixes it.

• Same rows and words: it returns { items, next_cursor: null } through pullRunLines, the same read function as GET …/app/pulls. limit is a whole number from 1 to 100, default 20, and anything else is refused with a sentence naming the range.
• Readers only: a stranger to a private repository gets not_found, and so does a repository with no app.
• Where it lives: the code is in src/http/mcp/tools/apps.ts, and docs/ADAPTERS.md now says so.
• Test: test/mcp-app-pulls.test.ts goes through the /mcp endpoint. It checks the result equals the API body for the same repo, that limit works, that a bad limit is refused, and that a stranger and a missing app get not_found.

I added the action to the existing apps tool rather than a new tool, and I did not raise any size ceilings. I did not add a pagination cursor, since the API doesn't have one either.

Closes #10102

MCP apps now reads an adapter's recorded pull runs: call it with { repo: "owner/name", pulls: true, limit? }. The new test, the type check, test/mcp-size.test.ts and test/mcp-apps-archived.test.ts pass. I ran bun scripts/gate.ts but didn't read its output, so treat that as unchecked. My first push left a type error in the test; the second push fixes it.

• Same rows and words: it returns { items, next_cursor: null } through pullRunLines, the same read function as GET …/app/pulls. limit is a whole number from 1 to 100, default 20, and anything else is refused with a sentence naming the range.
• Readers only: a stranger to a private repository gets not_found, and so does a repository with no app.
• Where it lives: the code is in src/http/mcp/tools/apps.ts, and docs/ADAPTERS.md now says so.
• Test: test/mcp-app-pulls.test.ts goes through the /mcp endpoint. It checks the result equals the API body for the same repo, that limit works, that a bad limit is refused, and that a stranger and a missing app get not_found.

I added the action to the existing apps tool rather than a new tool, and I did not raise any size ceilings. I did not add a pagination cursor, since the API doesn't have one either.

The check failed because no commit message had a "Try:" line. I added an empty commit with one and pushed it. The code is the same as the last push.

bun scripts/gate.ts printed nothing when I ran it, so I'm reading that as passing. I did not re-run the tests, since nothing but the commit message changed.

What changed: MCP apps now takes { repo: "owner/name", pulls: true, limit? } and returns an adapter's recorded pull runs, newest first. It reads them through pullRunLines, the same function behind GET …/app/pulls, so the rows and words match the API. The code is in src/http/mcp/tools/apps.ts, and docs/ADAPTERS.md says so.

Access: only readers get the rows. A stranger to a private repository, or a repository with no app, gets not_found. limit is a whole number from 1 to 100 (default 20); anything else is refused with a sentence naming the range.

Not done: I added an apps argument instead of a new tool, and I did not raise any size ceilings. I did not add a pagination cursor, since the API has none either.

Try it: on the preview's /mcp, call tool apps with {"repo":"owner/name","pulls":true} and compare it with GET /api/v1/repos/owner/name/app/pulls.

test/mcp-app-pulls.test.ts calls it through /mcp. It checks the result equals the API body, that limit works, and that a stranger and a missing app get not_found.