What problem does it solve? MCP servers suffer from slow cold starts, per-request connection overhead, and linear tool lookups that degrade response times as tool counts grow. This Skill provides implementation patterns to reduce latency and resource usage in claude-flow v3 MCP servers. ## Core Features & Use Cases - Connection Pooling: Reuse MCP connections with health checks, idle eviction, and pre-warming to eliminate per-request connection overhead. - Fast Tool Registry: Replace O(n) linear tool search with hash-based O(1) lookup, LRU caching, and fuzzy name matching across 200+ tools. - Load Balancing & Transport Optimization: Distribute requests across server instances using least-connections or response-time strategies, with message batching and compression. - Performance Monitoring: Collect p95 latency, error rates, pool hit rates, and health status with alerting thresholds. - Use Case: A developer notices their MCP server takes 1.8s to start and tool calls exceed 200ms; they apply these patterns to reach sub-400ms startup and sub-100ms p95 responses. ## Quick Start Ask the AI to analyze the current MCP server performance bottlenecks and implement connection pooling, load balancing, and tool registry optimization using an mcp-specialist agent.