ComfyUI_MCP Server（ComfyUI的ModelContextProtocol 服务端 | ModelConte…

What is ComfyUI_MCP Server?

ComfyUI_MCP Server is a loosely coupled, extensible, and configuration-driven Model Context Protocol (MCP) server designed for ComfyUI. It supports extending MCP services (tools) such as txt2img and img2img based on user-customized workflows, making it ideal for automation and integration in AI image generation and inference scenarios.

How to use ComfyUI_MCP Server?

Install by running python install.py, then start the server using run_mcp.bat (Windows) or run_mcp.sh (Linux). The server runs in streamable-http mode by default and automatically registers all tool modules placed in the tools/ directory. Configuration for ComfyUI address and logging is managed via config.ini.

Key features of ComfyUI_MCP Server

• Loosely coupled and configuration-driven MCP server for ComfyUI
• Auto-registers tools from the tools/ directory using @mcp.tool() decorator
• Config-driven parameters via JSON templates with default values
• Supports resource registration via @mcp.resource() decorator
• Default streamable-http transport; configurable in config.ini
• Extensible with custom tools and workflows without modifying the main entry

Use cases of ComfyUI_MCP Server

• Automate text-to-image generation with custom ComfyUI workflows
• Integrate image-to-image processing into external applications via MCP
• Extend MCP tools for custom AI inference pipelines
• Debug and test MCP tools interactively using MCP Inspector
• Manage and discover ComfyUI model resources through resource APIs

FAQ from ComfyUI_MCP Server

How do I add a new MCP tool?

Add a .py file in the tools/ directory implementing a register_xxx_tool(mcp) function and a corresponding _api.json file defining its parameter template. The tool is auto-registered without modifying the main server entry.

What are the runtime dependencies and requirements?

Python environment with dependencies from pyproject.toml (install via uv pip install -r pyproject.toml). A running ComfyUI instance must be accessible and configured in config.ini.

Where does generated data live?

Generated images are returned as Markdown image URLs from the ComfyUI HTTP API. Workflow JSON templates are stored in the workflows/ directory. Tool configurations reside in the tools/ directory.

What are known limitations?

The project acknowledges several planned improvements: richer error handling and logging, hot-reload of tools and configuration, multi-ComfyUI backend support, multimedia data interaction, and more comprehensive ComfyUI HTTP API integration.

What transport and authentication are supported?

Default transport is streamable-http; the transport parameter can be changed in config.ini. No authentication mechanism is mentioned in the README.