提交

Skills Mcp

@Jignesh-Ponamwar

A semantic skills registry over MCP that teaches AI agents how to do things correctly using curated expert skills that agents discover and load at the moment they need them. You can build your own. At the moment it includes 30 expert runbooks covering Stripe, Docker, FastAPI, React, Next.js, Terraform, Claude/OpenAI/Gemini APIs, and more, retrieved on demand via vector search.
概览

skills-mcp

skills-mcp is a self-hostable registry of expert AI agent skills, delivered over MCP (Model Context Protocol).

The core idea: instead of hoping an AI agent remembers the right way to set up Stripe webhooks, containerize a Python app, or write a structured code review — you give it a searchable library of curated, step-by-step procedures it can pull from on demand.

How it works

An agent calls skills_find_relevant("set up Stripe subscriptions with webhooks"), gets back a ranked list of matches with similarity scores, then loads the full instructions for the best match with skills_get_body. The instructions are expert-written runbooks — the right API patterns, security checklists, common mistakes, and go-live steps — not generic advice.

Discovery uses semantic vector search (Cloudflare Workers AI, bge-small-en-v1.5) so natural language queries match correctly even when phrasing varies. Only the short description and trigger phrases are embedded, keeping the search space clean. Full instructions are fetched separately, on demand.

Architecture

The server runs as a single Cloudflare Python Worker backed by Qdrant Cloud for vector storage. No separate backend, no GPU, no database server to manage. A seed script populates six Qdrant collections from SKILL.md files in the repository — one collection per disclosure tier (frontmatter, body, options, references, scripts, assets).

Agents follow a three-tier progressive disclosure pattern: discover → load body → fetch supplementary files only if the instructions reference them. Nothing is loaded speculatively.

What's included

30 bundled skills covering Stripe, Supabase, FastAPI, React, Next.js, Docker, GitHub Actions, Terraform, Cloudflare Workers, the Anthropic/OpenAI/Gemini APIs, PDF/Excel/Word/PowerPoint generation, SQL, web scraping, code review, test writing, MCP server building, and more.

Self-hosting

Under 10 minutes from clone to deployed Worker. Requires a free Qdrant Cloud cluster, a free Cloudflare Workers account, and Cloudflare Workers AI credentials for embeddings. A Docker path is also available for fully local use without Cloudflare.

服务器配置

{
  "mcpServers": {
    "skill-mcp": {
      "transport": "streamable-http",
      "url": "https://skill-mcp.jigneshponamwar.workers.dev/mcp"
    }
  }
}
© 2025 MCP.so. All rights reserved.

Build with ShipAny.