mcp-discovery-skill-forge

name	mcp-discovery-skill-forge
description	Discovers MCP servers from the MCP Registry API, evaluates availability and fit, guides Cursor installation, and scaffolds framework-specific skills from live docs. Use when user asks to find/install MCP servers, check what MCPs are available, or generate skills for frameworks like TanStack Start.

MCP Discovery + Skill Forge

Quick start

1. Capture user intent:
   • Find MCP servers
   • Install MCP server(s) in Cursor
   • Generate a new framework skill from docs
2. Run bun .agents/skills/mcp-discovery-skill-forge/scripts/registry-search.ts --search "<keyword>" --limit 30.
3. Return ranked MCP options with install notes and why each is relevant.

Workflow A: Discover MCPs from API

1. Query MCP Registry API via the helper script.
2. Normalize and rank matches by:
   • Name match with user query
   • Description relevance
   • Recency (updated_at if provided)
   • Ecosystem fit with current repo stack
3. Return:
   • Available now
   • Possibly relevant alternatives
   • Not found / unavailable

Workflow B: Install in Cursor

1. Confirm install scope:
   • User-level Cursor config, or
   • Project-level config
2. Check existing enabled MCP servers from local MCP descriptor folders before adding duplicates.
3. Provide exact install steps for Cursor and include required env vars/secrets.
4. Verify by listing/reading the MCP tool descriptors after install.

If installation requires auth, run the server's auth tool first (mcp_auth) and then re-verify availability.

Workflow C: Generate framework skill from docs

Use this when user says things like "create a skill for TanStack Start".

1. Fetch current docs with Context7 CLI:
   • npx ctx7@latest library "<official name>" "<full user request>"
   • npx ctx7@latest docs "<library-id>" "<full user request>"
2. Extract:
   • Core concepts
   • Setup commands
   • Common workflows
   • Gotchas/migration details
3. Scaffold a new skill directory in .agents/skills/<skill-name>/ with:
   • SKILL.md (short, actionable)
   • Optional REFERENCE.md for deeper details
   • Optional scripts when deterministic steps are repeated
4. Keep generated skill instructions implementation-oriented and current-docs-backed.

Output contract

For MCP discovery requests, always return:

• Query parameters used
• Top matches (name, why relevant, install readiness)
• Any unavailable/missing MCPs
• Recommended next action

For framework-skill generation requests, always return:

• Docs source used
• Scope covered in generated skill
• File paths created/updated
• Suggested validation command(s)