An MCP server for AI image generation, providing Claude Code and other MCP-compatible clients access to Google Gemini (Imagen) and OpenAI (DALL-E / GPT Image) models via simple tool calls.

Generate images using Google's Gemini / Imagen models. Defaults to Nano Banana 2 (gemini-3.1-flash-image-preview), Google's latest image model with 4K support, character consistency, and thinking mode.

Generate images using OpenAI's GPT Image models.

git clone https://github.com/looptech-ai/neoimage.git
cd neoimage

python3 -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt

You need at least one of:

• Google API key (for Gemini) — aistudio.google.com/apikey
• OpenAI API key (for GPT Image) — platform.openai.com/api-keys

Copy .mcp.json.example to .mcp.json in the project directory where you want to use neoimage, then fill in your paths and keys:

cp .mcp.json.example /path/to/your/project/.mcp.json

Edit the file:

{
  "mcpServers": {
    "neoimage": {
      "command": "/absolute/path/to/neoimage/venv/bin/fastmcp",
      "args": ["run", "/absolute/path/to/neoimage/server.py"],
      "env": {
        "GOOGLE_API_KEY": "your-google-api-key",
        "OPENAI_API_KEY": "your-openai-api-key"
      }
    }
  }
}

Paths must be absolute. Replace /absolute/path/to/neoimage with the actual path where you cloned this repo (e.g. /Users/yourname/dev/neoimage).

In Claude Code, you should see neoimage listed as an available MCP server. Try:

Generate a 16:9 dark slide background at 2K resolution

Generate a professional dark presentation slide on a #262626 background
titled "Q1 Results" with a bar chart showing revenue growth.
Use gemini-3-pro-image-preview, aspect ratio 16:9,
save to /tmp/slide-q1.png

• Best all-round: gemini-3.1-flash-image-preview (Nano Banana 2, default) — fast + high quality
• Presentation slides: aspect_ratio: 16:9 with image_size: 2K or 4K
• Complex compositions: Add thinking_level: high for better instruction following
• Quick drafts: image_size: 512px for fastest generation (~3-8s)
• Ultra-wide banners: aspect_ratio: 8:1 or 4:1 (NB2 only)
• Logos / icons with transparency: generate_image_openai with background: transparent, format: png
• High-detail illustrations: gpt-image-1.5 with quality: high
• Generated images save to the current working directory unless you specify output_path

• Never commit .mcp.json — it's in .gitignore and contains your API keys
• Use .mcp.json.example as a template (safe to commit — contains no real keys)
• Each team member uses their own API keys

• FastMCP — Python MCP server framework
• Google Generative Language API — Gemini / Imagen
• OpenAI Python SDK — GPT Image