Skip to main content
Glama

ComfyUI MCP Server

1. Overview

  • A server implementation for integrating ComfyUI with MCP.

  • ⚠️ IMPORTANT: This server requires a running ComfyUI server.

    • You must either host your own ComfyUI server,

    • or have access to an existing ComfyUI server address.


Related MCP server: ComfyUI MCP

2. Debugging

2.1 ComfyUI Debugging

python src/test_comfyui.py

2.2 MCP Debugging

mcp dev src/server.py

3. Installation and Configuration

3.1 ComfyUI Configuration

  • Edit src/.env to set ComfyUI host and port:

    COMFYUI_HOST=localhost
    COMFYUI_PORT=8188

3.2 Adding Custom Workflows

  • To add new tools, place your workflow JSON files in the workflows directory and declare them as new tools in the system.


4. Built-in Tools

  • text_to_image

    • Returns only the URL of the generated image.

    • To get the actual image:

      • Use the download_image tool, or

      • Access the URL directly in your browser.

  • download_image

    • Downloads images generated by other tools (like text_to_image) using the image URL.

  • run_workflow_with_file

    • Run a workflow by providing the path to a workflow JSON file.

      # You should ask to agent like this.
      Run comfyui workflow with text_to_image.json
    • example image of CursorAI

  • run_workflow_with_json

    • Run a workflow by providing the workflow JSON data directly.

      # You should ask to agent like this.
      Run comfyui workflow with this 
      {
        "3": {
            "inputs": {
                "seed": 156680208700286,
                "steps": 20,
          ... (workflow JSON example)
      }

5. How to Run

  • Example mcp.json:

    {
      "mcpServers": {
        "comfyui": {
          "command": "uv",
          "args": [
            "--directory",
            "PATH/MCP/comfyui",
            "run",
            "--with",
            "mcp",
            "--with",
            "websocket-client",
            "--with",
            "python-dotenv",
            "mcp",
            "run",
            "src/server.py:mcp"
          ]
        }
      }
    }

5.2 Using Docker

  • Downloading images to a local folder with download_image may be difficult since the Docker container does not share the host filesystem.

  • When using Docker, consider:

    1. Set RETURN_URL=false in .env to receive image data as bytes.

    2. Set COMFYUI_HOST in .env to the appropriate address (e.g., host.docker.internal or your server's IP).

    3. Note: Large image payloads may exceed response limits when using binary data.

5.2.1 Build Docker Image

# First build image
docker image build -t mcp/comfyui .
{
  "mcpServers": {
    "comfyui": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-p",
        "3001:3000",
        "mcp/comfyui"
      ]
    }
  }
}

5.2.2 Using Existing Images

Also you can use prebuilt image.

{
  "mcpServers": {
    "comfyui": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-p",
        "3001:3000",
        "overseer66/mcp-comfyui"
      ]
    }
  }
}

5.2.3 Using SSE Transport

  1. Run the SSE server with Docker:

    docker run -i --rm -p 8001:8000 overseer66/mcp-comfyui-sse
  2. Configure mcp.json (change localhost to your IP or domain if needed):

    {
      "mcpServers": {
        "comfyui": {
          "url": "http://localhost:8001/sse" 
        }
      }
    }

NOTE: When adding new workflows as tools, you need to rebuild and redeploy the Docker images to make them available.


Available Tools

4 tools
download_imageA

Download an image from a URL and save it to a file.

Args:
    url: The URL of the image to download.
    save_path: The absolute path to save the image to. Must be an absolute path, otherwise the image will be saved relative to the server location.

Inputs: url (required), save_path (required).

run_workflow_from_fileC

Run a workflow from a file.

Args:
    file_path: The absolute path to the file to run.

Inputs: file_path (required).

run_workflow_from_jsonC

Run a workflow from a JSON data.

Args:
    json_data: The JSON data to run.

Inputs: json_data (required).

text_to_imageB

Generate an image from a prompt.

Args:
    prompt: The prompt to generate the image from.
    seed: The seed to use for the image generation.
    steps: The number of steps to use for the image generation.
    cfg: The CFG scale to use for the image generation.
    denoise: The denoise strength to use for the image generation.

Inputs: prompt (required), seed (required), steps (required), cfg (required), +1 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.2/5.0

    Scored across 4 tools

    Disambiguation5/5

    Each tool has a clearly distinct purpose: downloading images, running workflows from files, running workflows from JSON, and generating images from text. There is no overlap in functionality, making it easy for an agent to select the correct tool for any given task without confusion.

    Naming Consistency5/5

    All tool names follow a consistent verb_noun pattern with underscores (e.g., download_image, run_workflow_from_file, run_workflow_from_json, text_to_image). This predictable naming scheme enhances readability and usability for agents.

    Tool Count4/5

    With 4 tools, the server is well-scoped for basic ComfyUI operations, covering key areas like image handling and workflow execution. However, it might feel slightly thin for advanced use cases, such as missing tools for managing workflows or querying server status, but it remains reasonable for its purpose.

    Completeness3/5

    The tools cover core functionalities like image generation and workflow execution, but there are notable gaps. For example, there are no tools for listing available workflows, updating or deleting workflows, or handling errors, which could limit agents in more complex scenarios within the ComfyUI domain.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers