MCP server exposing a local (or LAN) ComfyUI instance's HTTP API as tools, so an LLM client can queue generations, inspect the queue/history, upload reference images, browse installed models/nodes, and pull back generated images.
