Mcp Patterns

Comprehensive reference for building production MCP servers beyond the initial scaffold. Covers the full stack: OAuth 2.1 authentication, tool composition patterns (pipeline vs parallel), resource caching with TTL and LRU eviction, server-initiated user input via elicitation, sampling for agent loops, and interactive UIs through MCP Apps. The security rules alone are worth it: prompt injection defense, zero-trust allowlists, and token validation patterns that prevent the common mistakes. Built against the 2025-11-25 spec under the new Agentic AI Foundation governance. Pair it with Anthropic's mcp-builder for scaffolding, then use this for hardening and advanced features. The decision tree up front saves you from reading all 14 rules when you just need to add caching or debug a transport issue.

Install to Claude Code

npx -y skills add yonatangross/orchestkit --skill mcp-patterns --agent claude-code