Overview
Create MCP servers that actually help LLMs finish real tasks. Covers tool design, error messaging, and parameter trade-offs — keeping you from shipping MCPs that only look functional.
Features
- Guides for both Python FastMCP and Node MCP SDK
- Best practices for tool names and parameter design
- Error-message patterns that reduce LLM retry loops
- Real-world tasks for benchmarking server quality
- Publishing and distribution workflow
Try prompts like
Say these in XClaw — the skill activates automatically
- ›Scaffold a FastMCP server that wraps our internal HR API
- ›Package this REST API as a Node MCP server
- ›Review this MCP's tool design and flag spots the LLM is likely to get stuck
