MCP Builder

A comprehensive guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. MCP servers provide tools that allow LLMs to access external services and APIs. This skill covers both Python (FastMCP) and Node/TypeScript (MCP SDK) implementations, focusing on agent-centric design principles and evaluation-driven development.

Development follows a structured four-phase process:

Phase 1: Deep Research and Planning

• Study agent-centric design principles (build for workflows, not just API endpoints)
• Fetch and study MCP protocol documentation
• Load framework documentation (Python SDK or TypeScript SDK)
• Exhaustively study target API documentation
• Create comprehensive implementation plan

Phase 2: Implementation

• Set up proper project structure
• Implement core infrastructure (API helpers, error handling, formatting)
• Implement tools systematically with proper input validation (Pydantic/Zod)
• Follow language-specific best practices
• Add tool annotations (readOnlyHint, destructiveHint, etc.)

Phase 3: Review and Refine

• Review code quality (DRY principle, composability, consistency)
• Test and build (avoiding main process hangs)
• Use quality checklists from language guides

Phase 4: Create Evaluations

• Write 10 complex, realistic evaluation questions
• Ensure questions are independent, read-only, and verifiable
• Create XML evaluation file for testing