Comfy MCP Server

A server using FastMCP framework to generate images based on prompts via a remote Comfy server.

Overview

This script sets up a server using the FastMCP framework to generate images based on prompts using a specified workflow. It interacts with a remote Comfy server to submit prompts and retrieve generated images.

Prerequisites

• uv package and project manager for Python.

• Workflow file exported from Comfy UI. This code includes a sample Flux-Dev-ComfyUI-Workflow.json which is only used here as reference. You will need to export from your workflow and set the environment variables accordingly.

You can install the required packages for local development:

Configuration

Set the following environment variables:

• COMFY_URL to point to your Comfy server URL.

• COMFY_WORKFLOW_JSON_FILE to point to the absolute path of the API export json file for the comfyui workflow.

• PROMPT_NODE_ID to the id of the text prompt node.

• OUTPUT_NODE_ID to the id of the output node with the final image.

• OUTPUT_MODE to either url or file to select desired output.

Optionally, if you have an Ollama server running, you can connect to it for prompt generation.

• OLLAMA_API_BASE to the url where ollama is running.

• PROMPT_LLM to the name of the model hosted on ollama for prompt generation.

Example:

Usage

Comfy MCP Server can be launched by the following command:

Example Claude Desktop Config

Functionality

generate_image(prompt: str, ctx: Context) -> Image | str

This function generates an image using a specified prompt. It follows these steps:

1. Checks if all the environment variable are set.

2. Loads a prompt template from a JSON file.

3. Submits the prompt to the Comfy server.

4. Polls the server for the status of the prompt processing.

5. Retrieves and returns the generated image once it's ready.

generate_prompt(topic: str, ctx: Context) -> str

This function generates a comprehensive image generation prompt from specified topic.

Dependencies

• mcp: For setting up the FastMCP server.

• json: For handling JSON data.

• urllib: For making HTTP requests.

• time: For adding delays in polling.

• os: For accessing environment variables.

• langchain: For creating simple LLM Prompt chain to generate image generation prompt from topic.

• langchain-ollama: For ollama specific modules for LangChain.

License

This project is licensed under the MIT License - see the LICENSE file for details.