Tell the agent about the connected apps in the system prompt

[guilimasp]

When a Composio key is configured the harness mounts the Connect MCP server and pre-allows its tools, but the persona never mentions it — the computer and agents integrations each get a line, Composio gets none.

The result is that the model does not link a request like "can you read my email?" to the mcp__composio__* tools it is holding. It answers that it has no access, or starts hunting for an unrelated integration to authorize, while a working Gmail connection sits one tool call away.

Asked the same question before and after this line: before, the bot said it had no access and suggested authorizing a different connector; after, it searched the toolkits, ran the tool and answered from the mailbox.

One line, in the same shape as the two beside it. pnpm typecheck && pnpm test pass.

Summary by CodeRabbit

• New Features

  • Connected Composio apps are now available through supported providers as HTTP MCP integrations.
  • Supported providers advertise whether Composio integrations are available.
  • Connected-app guidance is shown only when the integration is active.

• Bug Fixes

  • Prevented Composio integrations from being mounted for providers that do not support them.

• Tests

  • Added coverage for integration mounting, authorization, API key forwarding, and capability reporting.

[milind-soni]

The intent is good, but this hint is currently gated only on integrations.composio. startTurn sets that integration for every provider whenever a key exists, while server/drivers/claude.ts is the only driver that mounts the Composio MCP server; Codex, Grok, and ACP drivers ignore it. That means this change would tell non-Claude bots to call tools they do not have. Please gate both the integration/prompt through an explicit capability (preferred), or at minimum restrict this prompt to the Claude driver until other drivers mount Composio.

[coderabbitai]

Warning

Review limit reached

@milind-soni, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 34 minutes

Limit details: You’ve used all 3 included reviews currently available under your plan.

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 39f2a57b-be00-4c20-bc16-921729b9f7cd

📥 Commits

Reviewing files that changed from the base of the PR and between 595be3a and ca4ac9e.

📒 Files selected for processing (3)

• server/drivers/acp/acp.test.ts
• server/drivers/claude.ts
• server/index.ts

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b6f5fc30-2247-46a7-9cf7-31efd01d838f

📥 Commits

Reviewing files that changed from the base of the PR and between 13a1bb7 and 595be3a.

📒 Files selected for processing (5)

• server/contracts.ts
• server/drivers/acp/acp.test.ts
• server/drivers/claude.test.ts
• server/drivers/claude.ts
• server/index.ts

---

📝 Walkthrough

Walkthrough

The provider capability contract now identifies Composio MCP support. The Claude driver enables this capability, ACP coverage confirms it remains disabled, and server wiring gates Composio credentials and prompt text on the selected adapter’s capability.

Changes

Composio MCP capability gating

Layer / File(s)	Summary
Capability contract and driver metadata server/contracts.ts, server/drivers/claude.ts, server/drivers/claude.test.ts, server/drivers/acp/acp.test.ts	The adapter contract adds composioMcp. Claude enables the capability and tests verify its MCP configuration. ACP tests verify that unsupported drivers do not enable it.
Conditional Composio integration wiring server/index.ts	Composio credentials and the system-prompt hint are added only when the selected adapter supports Composio MCP.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: ⚪ Minimal · up to 595be

This localized prompt change makes the existing connected-app tools discoverable to the agent without altering permissions or runtime behavior; no actionable merge-blocking risk remains after normal checks and review.

Suggested reviewers: milind-soni, clewkord, aivsomkar

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly describes the primary change: informing the agent about connected apps in the system prompt.
Description check	✅ Passed	The description explains what changed, why it changed, and how it was verified; omitted template headings and checklist items are non-critical.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[guilimasp]

You are right, and thank you — gating on cfg.composio.key would have told Codex, Grok and every ACP bot to call tools they were never handed. Took the preferred route.

• New capabilities.composioMcp, documented next to agentsMcp and computerMcp and following the same rule: the flag means this driver mounts the integration, not the user has connections.
• claude.ts declares it (it is the only driver that mounts the Connect MCP server today). Nothing else does.
• startTurn now gates integrations.composio on the capability, so a non-Claude bot gets neither the tools nor the hint. The prompt line hangs off integrations.composio rather than the key, the way the agents hint already hangs off integrations.agents — so the two can never drift apart.

Tests: claude.test.ts asserts it both mounts the server (pre-allowing mcp__composio) and claims the capability; acp.test.ts asserts ACP does not claim what it cannot honour. Any future driver that mounts Composio flips one flag and picks up the prompt with it.

pnpm typecheck && pnpm test pass. Rebased onto current main.

[milind-soni]

The requested provider-capability gating is now implemented: Composio configuration and prompt guidance are exposed only through adapters that declare composioMcp support. I refreshed the branch on current main while preserving both ACP/Droid test coverage, then verified the full test/build suite and the hosted macOS, Ubuntu, Windows, and Linux package-smoke matrix.

[milind-soni]

Heads-up after this merged: the prompt line named COMPOSIO_EXECUTE_TOOL, which doesn't exist on the endpoint the driver mounts. I checked tools/list against connect.composio.dev/mcp with the repo's own key — there are seven tools, and calling that name returns -32602: Tool COMPOSIO_EXECUTE_TOOL not found. The executor is COMPOSIO_MULTI_EXECUTE_TOOL, and arguments come from COMPOSIO_GET_TOOL_SCHEMAS, so search→execute alone can't build a valid call.

Fixed in #126 (merged): the hint now says SEARCH_TOOLS → GET_TOOL_SCHEMAS → MULTI_EXECUTE_TOOL. The idea was right — a bot that doesn't know it has connected apps tells the user it has no access — it just needed the real tool names.