Banana Image

Banana Image MCP is a production-ready MCP server that enables Claude and other AI assistants to generate high-quality images using Google's Gemini models. It delivers 4K output, smart model selection, grounding, and natural language editing, all with a simple, zero-setup workflow.

How to use

Describe what you want, and the Banana Image MCP server will generate the image through Claude Desktop. You’ll benefit from 4K output, smart model selection, and optional Google Search grounding for accurate results.

To enable the MCP server in Claude Desktop, add the following configuration to your Claude Desktop config file. This creates a banana_image MCP entry that runs locally via uvx.

{
  "mcpServers": {
    "banana-image": {
      "command": "uvx",
      "args": ["banana-image-mcp"],
      "env": {
        "GEMINI_API_KEY": "your-api-key-here"
      }
    }
  }
}

How to install

Prerequisites: you need a Gemini API key and Claude Desktop installed.

Install steps are designed to be quick and straightforward. Follow these commands in order to add Banana Image MCP to Claude Desktop.

# Install the MCP client environment if required
# This example follows the local, stdio-based setup described in the guide

Additional configuration and usage notes

Environment variable you must provide: GEMINI_API_KEY. Optionally, you can set IMAGE_OUTPUT_DIR to choose where generated images are saved. The default is ~/banana-images if you do not override it.

Update flow: when using uvx, clear the local package cache if you want to fetch the latest version, then restart Claude Desktop.

Model behavior and capabilities

The MCP server supports 4K output up to 3840px with the Pro model and faster 2–3s previews with the Flash model for quick iterations. It automatically selects the best model based on your prompt, and you can influence it with keywords like “quick sketch” for Flash or “4K professional” for Pro.

You can request different aspect ratios (for example 16:9, 1:1, 9:16, 4:3, 21:9) and ask for edits using natural language commands such as “make the sky more dramatic” or “remove the background.”

Environment and outputs

The server can generate multiple images per request (n between 1 and 4) and supports grounding for real-world accuracy when using the Pro model.

Images are saved to the output directory you specify with IMAGE_OUTPUT_DIR or the default ~/banana-images if you do not set it.

Troubleshooting and tips

If you don’t see results, verify that GEMINI_API_KEY is correct and that Claude Desktop has access to the MCP entry you configured. If you need to refresh the server, clear the uvx package cache and restart Claude Desktop.

Examples

Generate a 4K product photo of a coffee cup with studio lighting.

Create a 16:9 YouTube thumbnail for a cooking video with a vibrant, high-contrast look.

Edit an existing image to make the sky more dramatic.

Available tools

generate_image

Generates images from prompts using Gemini models with options for resolution, aspect ratio, and model tier.

edit_image

Edits an existing image using natural language commands.

grounding

Grounds results with Google Search for factual accuracy (Pro model).

model_selection

Smart selection of Flash or Pro model based on the prompt.