mcp-builder

Build an MCP server by designing workflow-oriented tools with JSON Schema validation, implementing secure input handling, and choosing a transport protocol—stdio, HTTP, SSE, or WebSocket. Follow the four-phase approach: research your service's API, implement tool handlers in TypeScript or Python, review for security and error handling, then evaluate with test cases.

Stdio servers run as subprocesses with stdin/stdout communication, ideal for local tools. HTTP servers handle request-response over standard web protocols. SSE servers use server-sent events for streaming. WebSocket servers enable bidirectional real-time communication. Choose based on your integration's latency, state management, and deployment requirements.

Yes. JSON Schema validation is required to define tool parameters, ensure type safety, and provide Claude with precise input constraints. This prevents malformed requests, improves error messages, and enables robust security-first input validation across all tool definitions.

MCP servers require comprehensive error handling with explicit error messages, configurable detail levels for debugging, and security-conscious input validation. Anticipate service failures, rate limits, and invalid requests; return structured errors that Claude can interpret and retry appropriately.

The skill supports both TypeScript and Python implementations independently. Choose one language for your server codebase to maintain consistency. However, you can reference examples in both languages to understand design patterns and apply them to your chosen implementation.

Workflow-oriented tools solve specific user problems by combining multiple API calls and business logic into cohesive actions, rather than exposing raw endpoints. They reduce Claude's decision complexity, handle authentication and state transparently, and deliver meaningful results with fewer round trips.