日本語版 (Read in Japanese)

Gemini Image MCP Server

This is an MCP (Model Context Protocol) server that uses Google's Gemini API to generate images and save them to a specified directory. In addition to text prompts, you can optionally provide input images to guide the image generation process. Generated images are automatically compressed to reduce file size.

Features

• Image generation from text prompts
• (Optional) Image generation using input reference images
• Automatic compression of generated images (JPEG, PNG)
• Unique file name assignment to prevent file name conflicts
• Operates as an MCP server, accepting tool calls via standard input/output

Prerequisites

• Node.js (v18 or higher recommended)
• Google Cloud Project with Gemini API enabled
• Gemini API Key

Setup

Example MCP server configuration for Roo Code

{
  "mcpServers": {
    "gemini-image-mcp-server": {
      "command": "npx",
      "args": [
        "-y",
        "@creating-cat/gemini-image-mcp-server"
      ],
      "env": {
        "GEMINI_API_KEY": "YOUR_GEMINI_API_KEY"
      },
      "disabled": false,
      "timeout": 300
    }
  }
}

• Replace YOUR_GEMINI_API_KEY with your actual Gemini API Key.

  • You can also use ${env:GEMINI_API_KEY} to retrieve the key from environment variables (Roo Code feature).

Tool: generate_image

This MCP server provides a tool named generate_image.

Input Parameters

Output

On success, the server returns the save path of the generated image and a message detailing the process, including the original and compressed file sizes. Example:

{
  "content": [
    {
      "type": "text",
      "text": "Image successfully generated and compressed at output/images/my_cat.jpg.\nOriginal size: 1024.12KB, Final size: 150.45KB"
    }
  ]
}

If an error occurs, an error message will be returned.

Notes

• The MIME type and aspect ratio of the generated images depend on the default settings of the Gemini API.
• Handle your API key with care.
• This server uses the model gemini-2.0-flash-preview-image-generation. Google may discontinue this model in the future.

License

MIT