ComfyUI MCP Server

Content

ComfyUI MCP Server is a source-install Model Context Protocol server that lets Claude control a local ComfyUI instance. It exposes tools for generating media, viewing generated images, regenerating assets with parameter overrides, polling jobs, listing assets, inspecting metadata, configuring defaults, running custom workflows, and publishing approved assets into project directories.

The project is aimed at local iterative media workflows. It can be useful for design exploration, web assets, prompt iteration, and custom ComfyUI workflow automation, but generated media still needs human review before it is shipped, published, or reused.

Source Review

• https://github.com/joenorton/comfyui-mcp-server
• https://github.com/joenorton/comfyui-mcp-server/blob/main/README.md
• https://github.com/joenorton/comfyui-mcp-server/blob/main/requirements.txt
• https://github.com/joenorton/comfyui-mcp-server/blob/main/server.py
• https://github.com/joenorton/comfyui-mcp-server/blob/main/comfyui_client.py
• https://github.com/joenorton/comfyui-mcp-server/blob/main/tools/generation.py
• https://github.com/joenorton/comfyui-mcp-server/blob/main/tools/publish.py
• https://github.com/joenorton/comfyui-mcp-server/blob/main/LICENSE

These sources were reviewed on 2026-06-06. Prefer the live repository, README, Python requirements, server implementation, ComfyUI client, generation tool, publish tool, and license file for current setup, transport options, tool behavior, asset handling, and licensing.

Features

• Source-install Python MCP server for local ComfyUI.
• Stdio mode available from server.py --stdio; upstream also documents local streamable HTTP usage.
• Workflow-backed generation tools such as generate_image, generate_song, regenerate, and custom workflow tools discovered from workflow JSON files.
• Asset tools for viewing images, listing recent assets, and reading metadata.
• Job tools for queue status, job polling, and cancellation.
• Configuration tools for defaults and model listing.
• Publish tools for copying, optimizing, and manifesting approved generated assets.
• Apache-2.0 license.

Installation

Clone the repository, install dependencies, start ComfyUI locally, and configure the MCP server from the source checkout:

git clone https://github.com/joenorton/comfyui-mcp-server
cd comfyui-mcp-server
pip install -r requirements.txt

{
  "mcpServers": {
    "comfyui-mcp-server": {
      "command": "python",
      "args": ["server.py", "--stdio"],
      "env": {
        "COMFYUI_URL": "",
        "COMFY_MCP_WORKFLOW_DIR": ""
      }
    }
  }
}

After restarting the MCP client, ask Claude to generate an approved asset, inspect the result, and iterate only after reviewing the output.

Use Cases

• Generate image concepts from approved prompts.
• Iterate on a generated asset without restating all parameters.
• Run custom ComfyUI workflows exposed as MCP tools.
• Ask Claude to list available models and defaults before generation.
• Poll or cancel long-running ComfyUI jobs.
• View generated image assets inside the client workflow.
• Publish reviewed assets into a web project directory with controlled filenames.

Safety and Privacy

ComfyUI MCP Server can create and publish generated media. Review prompts, models, workflow licenses, source inputs, and output rights before using assets in public or commercial contexts. Generated media can include unsafe, misleading, copyrighted, trademarked, biased, or otherwise sensitive content, even when prompts look harmless.

The server can store generated assets, metadata, workflow history, job IDs, publish paths, and manifest updates. Keep it connected only to trusted local clients, review publish targets before writing, and avoid sending private images, brand assets, customer materials, or regulated media through workflows unless your environment is approved for that data.

Duplicate Check

Existing MCP content includes image, browser, and design-related integrations, but no dedicated entry for joenorton/comfyui-mcp-server was found in content/mcp. This entry is distinct because it covers a local ComfyUI MCP bridge for generated media workflows, job management, asset inspection, custom workflow execution, and publish tools.