#What it does

The MCP Builder skill guides the creation of production-quality MCP (Model Context Protocol) servers. These servers enable LLMs to interact with external services through well-designed tools, following a structured four-phase development process from research to evaluation.

#How to use

Activate when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or TypeScript (MCP SDK).

Build an MCP server for the GitHub API

Create an MCP server that integrates with Slack using TypeScript

#Skill instructions

#Four-Phase Development Process

Phase 1: Deep Research and Planning

• Study MCP protocol documentation and specification
• Load framework documentation (TypeScript recommended for broad compatibility)
• Understand the target API's endpoints, authentication, and data models
• Plan tool selection prioritizing comprehensive API coverage

Phase 2: Implementation

• Set up project structure with proper configuration
• Implement core infrastructure (API client, error handling, pagination)
• Build tools with:
  • Input schemas using Zod (TypeScript) or Pydantic (Python)
  • Output schemas with structuredContent where possible
  • Clear descriptions for agent discoverability
  • Annotations (readOnlyHint, destructiveHint, idempotentHint)

Phase 3: Review and Test

• Code quality review (DRY, consistent error handling, full type coverage)
• Build verification and compilation checks
• Test with MCP Inspector

Phase 4: Create Evaluations

• Generate 10 complex, realistic test questions
• Verify answers using read-only operations
• Questions must be independent, stable, and verifiable
• Output as XML evaluation file

#Recommended Stack

• Language: TypeScript (best SDK support and AI model compatibility)
• Transport: Streamable HTTP for remote servers, stdio for local servers
• Key principle: Balance comprehensive API coverage with specialized workflow tools

This skill is from the Anthropic Skills Repository.