Image Gen MCP

🚀 AI Image Generation Server Based on MCP Interface

This project offers a Model Context Protocol (MCP) service integrating the image generation functionality of Stable Diffusion, allowing AI agents to request and receive generated images. This project is based on the example provided in Block's Goose Custom Extension Tutorial.

🚀 Quick Start

📦 Installation

1. Create a virtual environment and force it to use .venv:

   virtualenv .venv
   

2. Activate the virtual environment:

   source .venv/bin/activate
   

3. Install the MCP package (for Goose integration):

   pip install -e .
   

💻 Usage Examples

Starting the Service

The MCP server includes an integrated image generation service. You can start both with the following commands:

Standard Mode:

source .venv/bin/activate  # Activate your virtual environment
export IMAGE_GEN_DIR=/absolute/path/to/folder # Set the target folder for generated images
image-gen-mcp

Development Mode (with FastMCP Inspector): Open two terminals:

Terminal 1:

source .venv/bin/activate  # Activate your virtual environment
export IMAGE_GEN_DIR=/absolute/path/to/folder # Set the target folder for generated images
image-gen-mcp # Start the image generation service (and an MCP server we don't use)

Terminal 2:

source .venv/bin/activate  # Activate your virtual environment
export IMAGE_GEN_DIR=/absolute/path/to/folder # Set the target folder for generated images
python -m src.image_gen_mcp.server # Start the MCP server

⚠️ Important Note

• In development mode, the MCP server runs in src/image_gen_mcp/server.py.
• The image generation service and the MCP server run in the same process.

Stopping the Service

Use Ctrl+C to stop both services since they are now running in the same process.

📚 Documentation

Integration with Goose

Add the project executable path to uv run:

uv run /full/path/to/your/project/.venv/bin/image-gen-mcp

Local Running Recommendations

💡 Usage Tip

• It works best when using OpenAI GPT - 4o.
• If using Ollama, it is recommended to try the mistral - small3.1 model.
• Even on a high - end MacBook Pro, mistral - small3.1 is much slower than OpenAI GPT - 4o.

Service Architecture