An Agent Skill for building high-quality MCP servers with FastMCP.
Combines FastMCP framework guidance with battle-tested design patterns from Arcade to help AI agents build MCP servers that are well-structured, secure, and optimized for LLM consumption.
The skill covers 11 areas of MCP server development:
- Before you start — always fetch latest docs
- Guiding principles — machine experience, tool DAGs, error-guided recovery, security boundaries
- Tool design — classification, descriptions, input constraints, defaults, identifiers
- Resources & templates — when to use, URI design, discoverability
- Prompts — when and how to use reusable message templates
- Context & state — dependency injection, logging, progress, session state
- Tool output — response shaping, token efficiency, pagination, next-action hints
- Error handling — recovery-oriented errors, classification, ambiguity handling
- Server composition — mounting, providers, transforms, abstraction patterns
- Authentication & security — credential handling, authorization, audit, boundaries
- Anti-patterns — common mistakes that degrade agent effectiveness
npx skills add https://github.com/detailobsessed/wise-mcp --skill wise-mcpThis project uses prek for pre-commit hooks and uv for Python tooling.
# Install pre-commit hooks
prek install
# Run all checks
prek run --all-filesChecks include: trailing whitespace, EOF fixer, large file detection, case conflicts, merge conflicts, TOML/YAML validation, mixed line endings, typos spell checking, and lychee link validation.
MIT