docs/MCP.md names every ui:// card the server draws, and a test fails when a card is drawn but not described #11174

For: an agent or harness reading docs/MCP.md (the documentation /.well-known/mcp.json points to) learns every card the forge can show.

Problem today: the source draws ui://forge/issue, board, crew, flow and forks. The Cards, the contract and the plugins section lists only app, run, asks, release, wave, map, feed, pull, pulls, runs and asked, as one long sentence.

Done when:

• [x] The section is a short table, one row per card: resource URI, the tool that draws it, and one line on what the card shows.
• [x] A new test collects every ui://forge/<name> the server can return (from its resource list, not a grep of the docs). It fails when one has no row in docs/MCP.md, or a row names a card the server no longer draws.
• [x] bun test test/mcp-size.test.ts passes: no tool description grows, no ceiling is raised.

Files: docs/MCP.md, one new test. Risk: docs and test only; no tool changes.

Reviewed pull #11180 at its head: ticked The section is a short table, one row per card: resource URI, the tool that draws it, and one line on what the card shows.; A new test collects every ui://forge/<name> the server can return (from its resource list, not a grep of the docs). It fails when one has no row in docs/MCP.md, or a row names a card the server no longer draws.; bun test test/mcp-size.test.ts passes: no tool description grows, no ceiling is raised..