MCP restore_drill, scorecard, reports and usage draw ui:// cards, as held and board do #11503

restore_drill, reports and usage now draw ui://forge/drill, ui://forge/reports and ui://forge/usage. scorecard already had a ui://forge/scorecard card, so I changed its rows instead of adding one. The tools' JSON is unchanged and no tool was added.

The new tests, tsc, design-lint, mcp-size, mcp-cards-coverage, mcp-card-table, mcp.test.ts and mcp-held-tool-card pass. scripts/gate.ts is present and I ran it, but its result wasn't checked. I did not run the full suite.

• drill: the card shows the verdict as one word (green, red or none), the failed step when red, and when it ran. The tool passes the drill's at to the card as the held tool does for its repo, so the JSON doesn't change.
• scorecard: each agent's row now leads with all-time landed and red rate, then the week's figures. I read "the number beside its dial" as red rate; say if you meant something else.
• reports: one row per waiting report with its state and who filed it, each linked to the repository's /reports page. An accept or dismiss answer gets no card.
• usage: for a repo, a total row plus one row per agent. For the forge, rows for today and the last 24 hours, plus one per subscription with its headroom.
• docs and ceilings: docs/MCP.md lists the three new cards and has the scorecard row reworded. I ran bun scripts/mcp-table.ts, and no ceiling was raised. Each card tool's size in docs/MCP.md went up by about 50 bytes, from the card link in tools/list.
• tests: test/mcp-ops-cards.test.ts has one test per card. They call cardFor directly rather than going through a running server like test/mcp-held-tool-card.test.ts does. I also added the three new URIs to the expected list in test/mcp.test.ts.
• not done: no buttons on any of the four cards. The pre-push hook wanted a "how to try it" commit, so there is an empty Try: commit.

To try it, call any of the four tools from an MCP Apps host (an admin token for restore_drill, scorecard, and usage without a repo).

Closes #11475

restore_drill, reports and usage now draw ui://forge/drill, ui://forge/reports and ui://forge/usage. scorecard already had a ui://forge/scorecard card, so I changed its rows instead of adding one. The tools' JSON is unchanged and no tool was added.

The new tests, tsc, design-lint, mcp-size, mcp-cards-coverage, mcp-card-table, mcp.test.ts and mcp-held-tool-card pass. scripts/gate.ts is present and I ran it, but its result wasn't checked. I did not run the full suite.

• drill: the card shows the verdict as one word (green, red or none), the failed step when red, and when it ran. The tool passes the drill's at to the card as the held tool does for its repo, so the JSON doesn't change.
• scorecard: each agent's row now leads with all-time landed and red rate, then the week's figures. I read "the number beside its dial" as red rate; say if you meant something else.
• reports: one row per waiting report with its state and who filed it, each linked to the repository's /reports page. An accept or dismiss answer gets no card.
• usage: for a repo, a total row plus one row per agent. For the forge, rows for today and the last 24 hours, plus one per subscription with its headroom.
• docs and ceilings: docs/MCP.md lists the three new cards and has the scorecard row reworded. I ran bun scripts/mcp-table.ts, and no ceiling was raised. Each card tool's size in docs/MCP.md went up by about 50 bytes, from the card link in tools/list.
• tests: test/mcp-ops-cards.test.ts has one test per card. They call cardFor directly rather than going through a running server like test/mcp-held-tool-card.test.ts does. I also added the three new URIs to the expected list in test/mcp.test.ts.
• not done: no buttons on any of the four cards. The pre-push hook wanted a "how to try it" commit, so there is an empty Try: commit.

To try it, call any of the four tools from an MCP Apps host (an admin token for restore_drill, scorecard, and usage without a repo).

I fixed the failing scorecard test and pushed. The test was out of date, not the code.

An earlier commit on this pull made the scorecard card's agent rows start with landed count and red rate. The test in test/mcp-flow-card.test.ts still expected the old text. I changed line 108 to expect 1 landed · 100% red · 1 merged · 0 returned · $0.00 this week.

test/mcp-flow-card.test.ts passes (6 of 6) and tsc --noEmit is clean. I did not run scripts/gate.ts or the design lint, since this change only touches a test. The forge will run the full check on the push.

What the pull does: the MCP restore_drill, reports, usage and scorecard tools now draw ui:// cards beside their unchanged JSON, and the scorecard card leads with landed and red rate.

To try it, call those tools from a host that draws ui:// cards.

🤖 Generated with Claude Code

The MCP tools restore_drill, scorecard, reports and usage now answer with ui:// cards beside their unchanged JSON, as held and board do. This push adds the tool-side tests @sentinel asked for. The new tests, the existing card tests and tsc pass. I didn't run scripts/gate.ts; the push followed straight after the other checks.

• New test: test/mcp-ops-tool-cards.test.ts goes through tools/call with an admin token. It uses a stub drill service and a repo with two filed reports.
• restore_drill: it asserts the card has the Verdict row, the Failed step row and the When row (3h ago).
• reports: it asserts both list rows link to /alice/engine/reports.
• usage: it asserts a card appears for the repo answer (a Total row) and for the forge answer (Today and Last 24 hours rows).
• Checked it bites: with card: { at } and card: { repo } removed, the restore_drill and reports tests fail.

I made no change to the tools and no change to the README.

The check stopped on CPU use, not on a test failure. Its log shows the browser "looks" step (headless Chrome on /, /board and the other pages in forge.json) pushing the runner past 16 cores. This change touches only the MCP tool cards and their tests, and no rendered page. I couldn't reproduce the CPU overrun here, so I'm treating it as a load spike on the runner.

I made no code change. Locally, the gate and tsc --noEmit are clean, and the four affected test files pass (61 tests). I'm not pushing an empty commit; a re-run of the check on the existing push should settle it.

What changed on this pull:

• restore_drill, reports and usage now draw ui:// cards.
• The scorecard card leads with landed and red rate.
• The JSON output is unchanged.
• docs/MCP.md is updated.

What I chose not to do: change forge.json or the looks list to work around the CPU limit.

How to try it: call the MCP tools restore_drill, reports, usage and scorecard in a host that draws ui:// cards. Each shows rows beside the unchanged JSON.

I merged main into the branch and pushed the result (48f873a6f). The only conflict was in src/web/studio/cards.ts, where both sides added a card branch at the same spot. I kept both: my drill, reports and usage cards, and main's new forks/explore card, which uses the exploreTreeRows, forkChainRows and forksPageRows helpers.

• Gate and type check: bun scripts/gate.ts and bun x tsc --noEmit both ran clean.
• Tests: The four MCP card tests from my change pass (61 tests). The card tests that cover the merged area (mcp-card*, forks, explore, drill) also pass (60 tests).
• Full suite not run here: scripts/fast.ts exited 2 because scripts/unreached-allow.json changed on main, so the diff is too broad to name. I picked the covering tests myself, and the forge's check runs the full suite on this push.