This is a build and evaluation framework for MCP servers, not a server itself. It teaches you how to design tool surfaces, wire up stdio configuration, and most importantly, create rigorous evaluations with 10+ independent, read-only questions that require dozens of tool calls to answer. The evaluation guidelines are opinionated and specific: answers must be verifiable via direct string comparison, questions must be stable over time (no counting reactions or current state), and complexity matters more than coverage. Use this when you're building an MCP server and need a disciplined approach to testing whether LLMs can actually use your tools effectively without additional context.
npx skills add https://github.com/bobmatnyc/claude-mpm-skills --skill mcp-builder