# ComfyUI MCP

> ComfyUI MCP is an open-source MCP server for design and media. Self-hosted MCP bridge that lets agents generate images and video on your ComfyUI GPU. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** self-hosted, Linux, API
- **Repository:** https://github.com/germaniu/comfyui-mcp
- **MCP Registry name:** io.github.GermaniU/comfyui-mcp
- **Categories:** [design and media](https://ai-powered.com/categories/design-and-media), [AI models and LLM tooling](https://ai-powered.com/categories/ai-and-llm)
- **Listing updated:** 2026-11-18

## About

This is only the MCP and transport layer. The ComfyUI engine and model weights are not included, so you run ComfyUI yourself and point COMFYUI_URL at it. The server then builds the workflow JSON and serves it to clients on your network over HTTP/SSE (default port 8201), so client machines need no PyTorch or local SDXL models.

A built-in GPU arbiter switches a 12 GB card between llama-server and ComfyUI to avoid VRAM collisions. Image generation offers presets such as producto (RealVisXL V4.0), realista (Juggernaut XL), rapido and anime, with aspect ratios, seeds, batches of up to four, optional LoRA and a face-detail pass.

Video uses Wan 2.2 and needs separate models. The README reports local benchmarks of roughly 27-32 minutes for a 5-second 5B clip and about 8 minutes for an A14B image-to-video clip, and says the MCP video integration has not yet been tested against the real GPU. It also warns that without MCP_AUTH_TOKEN anyone on the LAN can use the GPU.

## Features

- **Text to image:** generate_image runs SDXL with presets, aspect ratios, seeds and batches of 1-4.
- **Image variations:** img2img varies a generated image with an adjustable denoise level.
- **Short video:** generate_video makes 1-5 second MP4 clips with Wan 2.2.
- **GPU arbiter:** Alternates a 12 GB GPU between llama-server and ComfyUI without VRAM clashes.
- **Optional bearer auth:** MCP_AUTH_TOKEN makes every HTTP request carry an Authorization header.
- **Preview thumbnails:** Can return a small JPEG so vision-capable models see the result.

## Good for

- Give Claude Code or Cursor an image generator on your own GPU
- Produce product-style images for an agent workflow
- Create variations of an earlier render
- Share one GPU between a local LLM and image generation

## Install

### Claude Code (Terminal)

```sh
claude mcp add comfyui-mcp --env MCP_AUTH_TOKEN='<MCP_AUTH_TOKEN>' -- docker run -i --rm -e MCP_AUTH_TOKEN ghcr.io/germaniu/comfyui-mcp:0.1.1
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "comfyui-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "MCP_AUTH_TOKEN",
        "ghcr.io/germaniu/comfyui-mcp:0.1.1"
      ],
      "env": {
        "MCP_AUTH_TOKEN": "<MCP_AUTH_TOKEN>"
      }
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "comfyui-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "MCP_AUTH_TOKEN",
        "ghcr.io/germaniu/comfyui-mcp:0.1.1"
      ],
      "env": {
        "MCP_AUTH_TOKEN": "<MCP_AUTH_TOKEN>"
      }
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "comfyui-mcp": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "MCP_AUTH_TOKEN",
        "ghcr.io/germaniu/comfyui-mcp:0.1.1"
      ],
      "env": {
        "MCP_AUTH_TOKEN": "<MCP_AUTH_TOKEN>"
      }
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "comfyui-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "MCP_AUTH_TOKEN",
        "ghcr.io/germaniu/comfyui-mcp:0.1.1"
      ],
      "env": {
        "MCP_AUTH_TOKEN": "<MCP_AUTH_TOKEN>"
      }
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "comfyui-mcp": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "MCP_AUTH_TOKEN",
        "ghcr.io/germaniu/comfyui-mcp:0.1.1"
      ],
      "env": {
        "MCP_AUTH_TOKEN": "<MCP_AUTH_TOKEN>"
      }
    }
  }
}
```

## Tools

- `generate_image`: Text-to-image with SDXL presets, aspect ratios and optional LoRA.
- `img2img`: Vary an existing generated image with a denoise level.
- `generate_video`: Make a 1-5 second MP4 with Wan 2.2.

## FAQ

### Does it include ComfyUI?

No. You need a separate ComfyUI backend and model files; this project only provides the MCP interface.

### What does it need to run?

Python 3.11+ and uv or pip; the registry also lists an OCI image.

### Is it secure on a LAN?

Only if you set MCP_AUTH_TOKEN. Without it, any LAN host can generate images and occupy the GPU.

### Is video generation proven?

The README says the MCP integration has not been tested against the real GPU yet, though it gives local benchmark timings.

### Is ComfyUI MCP free?

Yes. ComfyUI MCP is open source, so you can use and self-host it for free.

### How do I add ComfyUI MCP to Claude Code?

Run this command in your terminal: claude mcp add comfyui-mcp --env MCP_AUTH_TOKEN='<MCP_AUTH_TOKEN>' -- docker run -i --rm -e MCP_AUTH_TOKEN ghcr.io/germaniu/comfyui-mcp:0.1.1

### What are the best alternatives to ComfyUI MCP?

The closest alternatives to ComfyUI MCP are HyperFrames by HeyGen, headroom, omniget, next-ai-draw-io and mcp-for-blender.

## Where this comes from

- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [HyperFrames by HeyGen](https://ai-powered.com/mcp/hyperframes-by-heygen): HyperFrames by HeyGen is a freemium MCP server for design and media.
- [headroom](https://ai-powered.com/mcp/headroom-3): headroom is an open-source MCP server for ai models and llm tooling.
- [omniget](https://ai-powered.com/mcp/omniget): omniget is an open-source MCP server for ai models and llm tooling.
- [next-ai-draw-io](https://ai-powered.com/mcp/next-ai-draw-io-2): next-ai-draw-io is an open-source MCP server for ai models and llm tooling.
- [mcp-for-blender](https://ai-powered.com/mcp/mcp-for-blender): mcp-for-blender is an open-source MCP server for design and media.
- [ffmpeg-skill](https://ai-powered.com/mcp/ffmpeg-skill): ffmpeg-skill is an open-source MCP server for design and media.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [unity-mcp](https://ai-powered.com/mcp/unity-mcp-2): unity-mcp is an open-source MCP server for ai models and llm tooling.

---
Source: https://ai-powered.com/mcp/comfyui-mcp-2
