This server bridges ComfyUI's node-based workflow system into the Model Context Protocol, letting AI agents generate images and videos by calling your custom workflows as tools. It exposes operations like queue_prompt, mount_workflow, upload_assets, and save_task_assets, essentially wrapping ComfyUI's API format JSON into parameterized functions. You define which node parameters are exposed to the AI, preventing hallucinated inputs while keeping full control over the underlying graph. The server handles asset uploads from local paths or URLs, tracks generation progress, and downloads outputs automatically. It's a zero-modification layer that sits between your MCP client and an existing ComfyUI instance, turning your workflows into callable endpoints without touching ComfyUI itself. Supports both STDIO and HTTP transports with bilingual tooling.
claude mcp add --transport stdio io.github.metabrain-labs-comfyui-mcp-server uvx comfyui-mcp-server