Mcp Builder

This walks you through building MCP servers that let LLMs interact with external APIs, covering both TypeScript and Python implementations. It's opinionated about design choices: favor comprehensive API coverage over workflow shortcuts, use TypeScript with streamable HTTP for remote servers, and prioritize clear tool naming with consistent prefixes. The guide includes four phases from research through evaluation, with specific sections on studying the MCP spec, setting up error handling, and creating test questions that verify your server actually works. The evaluation section is notably detailed, requiring you to create ten complex, read-only questions with verifiable answers. Use this when you're integrating a third-party service and want to avoid common pitfalls around tool design and agent discoverability.

Install to Claude Code

npx -y skills add malob/nix-config --skill mcp-builder --agent claude-code