mcp-builder

Build an MCP server using FastMCP (Python) or the Node/TypeScript MCP SDK to expose external services as tools. The process covers planning, tool design, input validation with Pydantic or Zod, error handling, and evaluation to ensure LLMs can reliably access and use your APIs.

Design agent-centric tools with clear naming conventions, concise JSON or Markdown responses, robust input validation, actionable error messages, and pagination for large datasets. This ensures LLMs understand and execute tools reliably within limited context windows.

Yes, MCP servers work across Python (FastMCP) and Node/TypeScript SDKs. Choose based on your existing stack; both support the same core workflows for tool design, validation, error handling, and testing.

Implement input validation using Pydantic (Python) or Zod (TypeScript) to catch invalid requests early, and design clear, actionable error messages that guide LLM recovery. Include robust testing and documentation practices to catch edge cases before production.

Document tool metadata, response formats, and error conditions clearly. Test across real-world workflows from research to production, validate pagination and truncation strategies, and verify behavior under constraints to catch functional and technical issues early.

MCP servers require foundational API integration and LLM knowledge, but this Skill covers the full workflow from planning through evaluation. Implementation depth is advanced, addressing naming conventions, validation models, transport, and production deployment patterns.