Reduce noisy MCP missing-capability reports

[masnwilliams]

summary

• tell agents to report only genuine server capability gaps after checking available tools
• exclude existing fallbacks, transient failures, capacity limits, and client-side permission restrictions
• cover the updated tool contract in the MCP analytics integration test

why

The previous description explicitly encouraged reports even when an existing tool could complete the task. That inflated missing-capability analytics with fallback, capacity, and client-policy issues.

testing

• bun test
• bun test src/lib/mcp/analytics.test.ts
• bunx prettier --check src/lib/mcp/analytics.ts src/lib/mcp/analytics.test.ts

bun run format:check still reports the pre-existing AGENTS.md formatting warning.

---

Note

Low Risk
Agent-facing copy and test expectations only; no change to analytics capture, sanitization, or tool registration behavior beyond what agents read in tools/list.

Overview
Narrows when MCP agents should call get_more_tools so missing-capability analytics reflect real server gaps instead of fallback use, outages, or client policy blocks.

The tool description in registerMissingCapabilityTool no longer suggests reporting when an existing tool could work. It now requires checking listed tools first and directs agents away from this path for existing fallbacks, transient failures or capacity limits, and client-side permission restrictions—preferring normal tools or submit_feedback instead.

The SDK integration test was renamed to keeps analytics tool contracts stable and now asserts the get_more_tools description includes those guardrails alongside the existing feedback-tool schema checks.

^{Reviewed by Cursor Bugbot for commit 933bc1b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
mcp	Ready	Preview	Sep 1, 2026 5:49pm UTC

[dprevoznik]

Lgtm