When you're building an MCP server to connect external APIs to LLMs, this walks you through the full workflow: researching the protocol, choosing between comprehensive API coverage versus workflow-specific tools, implementing with TypeScript or Python, and creating evaluations to test if agents can actually use what you built. It prioritizes streamable HTTP for remote servers and pushes you toward TypeScript for better compatibility. The evaluation phase is solid, it forces you to write 10 complex, realistic questions and verify the answers yourself before claiming the integration works. Most helpful when you're past the "hello world" stage and need to make real decisions about tool naming, error messages, and whether to expose every endpoint or build higher-level operations.
npx skills add https://github.com/unclecatvn/agent-skills --skill mcp-builder