ck:mcp-builder

To build an MCP server for an external API, design workflow-first tools rather than thin API wrappers, implement validated schemas using FastMCP or MCP SDK, and create evaluation questions to verify reliability. This ensures LLM agents can complete real-world tasks effectively.

The best way to design LLM tools for MCP servers is building around end-to-end agent workflows. This involves creating composite operations that reduce unnecessary calls, applying schema validation, and formatting responses consistently in JSON or Markdown for reliable LLM consumption.

Yes, the MCP SDK supports both TypeScript and Python implementations. Specifically, it provides protocol-aligned implementation guidance for Node/TypeScript using MCP SDK and Python using FastMCP, ensuring proper infrastructure setup and systematic tool deployment.

Handle pagination and errors in MCP tools by applying character limits and returning actionable error messages. Designing LLM-friendly inputs and outputs requires consistent response formats and schema validation to prevent agent confusion during API integration.

Evaluate the effectiveness of MCP server tools by creating complex, read-only evaluation questions. These questions measure tool usefulness and reliability by testing how well LLMs can complete real-world tasks via the implemented tools under various constraints.