Developer Builds Python Proxy That Hot-Swaps MCP Servers Without Restarting Agents

A developer published an open-source tool called Smart MCP Proxy that allows AI agents to add or remove Model Context Protocol (MCP) servers at runtime without requiring a restart. The single Python process uses a shared subprocess pool, so multiple agents connecting to multiple servers reuse the same pools rather than spawning redundant subprocesses, reducing memory overhead. A file watcher monitors the configuration YAML and automatically diffs changes, closing old server pools and spinning up new ones on the fly. The proxy also includes an AI concierge layer that accepts plain-English queries, routes them to the appropriate server using MCP Sampling, and returns only the final answer, keeping intermediate tool calls out of the agent's context. The tool requires no database, Docker, or embedded API keys, and can be installed via pip.