Nano-Banana MCP (Gemini 3 Pro Image)

A Model Context Protocol (MCP) server for AI image generation and editing using Google's Gemini 3 Pro Image model. Supports 4K output, aspect ratio control, Google Search grounding, and conversational multi-turn editing with thought signatures.

Fork of ConechoAI/Nano-Banana-MCP, upgraded from Gemini 2.5 Flash to Gemini 3 Pro Image.

Features

• Generate images from text prompts with configurable aspect ratio and resolution (up to 4K)
• Edit existing images by providing a file path and edit instructions
• Conversational editing — multi-turn editing that preserves full context via Gemini 3 thought signatures
• Google Search grounding — generate fact-based images using real-time web data
• Cross-platform file management (macOS, Linux, Windows)

Quick Setup

Claude Code

claude mcp add nano-banana \
  -e GEMINI_API_KEY=your-api-key \
  -- npx -y nano-banana-mcp-g3

Manual (Claude Code / Cursor / other MCP clients)

Add to your MCP configuration:

{
  "nano-banana": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "nano-banana-mcp-g3"],
    "env": {
      "GEMINI_API_KEY": "your-api-key"
    }
  }
}

Restart your client after saving.

Package name: nano-banana-mcp-g3 (not nano-banana-mcp). The original nano-banana-mcp on npm is the upstream version using the older Gemini 2.5 model.

Available Tools

| Tool | Description | |------|-------------| | generate_image | Create a new image from a text prompt | | edit_image | Modify an existing image file with a text prompt | | continue_editing | Iterate on the last image (preserves context via thought signatures) | | get_last_image_info | Show file path and info for the last image | | configure_gemini_token | Set your Gemini API key | | get_configuration_status | Check if the API key is configured |

Generation Parameters

These optional parameters work with generate_image, edit_image, and continue_editing:

| Parameter | Values | Description | |-----------|--------|-------------| | aspectRatio | 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, 21:9 | Output aspect ratio. Default is model-chosen. | | imageSize | 1K, 2K, 4K | Output resolution. Default is 1K. | | useGoogleSearch | true / false | Enable Google Search grounding for fact-based images (generate_image only). |

Usage Examples

Ask your AI assistant naturally:

• "Generate an image of a sunset over mountains"
• "Generate a 16:9 widescreen banner for a tech blog in 2K resolution"
• "Generate an infographic of today's weather in Tokyo" (uses Google Search grounding)
• "Edit this image to add a rainbow" (provide file path)
• "Make the background darker" (continues editing the last image)
• "Now change the text to be gold colored" (multi-turn editing with context)

Conversational Editing

The continue_editing tool uses Gemini 3's thought signatures to maintain full context between turns. The model truly "remembers" the previous image — it doesn't just re-read the file. This enables coherent multi-turn editing:

Turn 1: "Generate a logo for Morning Ritual coffee shop"
Turn 2: "Make it dark background with gold text"       <- remembers the logo
Turn 3: "Add a coffee bean pattern in the background"   <- remembers logo + dark/gold scheme

Generated images are saved to ./generated_imgs/ in your current working directory.

Configuration Priority

The API key is loaded in this order:

1. MCP config environment variables — set via "env": { "GEMINI_API_KEY": "..." } in your MCP config
2. System environment variables — export GEMINI_API_KEY="..."
3. Local config file — created via the configure_gemini_token tool (stored as .nano-banana-config.json)

Proxy Support

If you need to route requests through a proxy (e.g., for geo-restricted regions), set the GOOGLE_GEMINI_BASE_URL environment variable:

{
  "nano-banana": {
    "type": "stdio",
    "command": "npx",
    "args": ["-y", "nano-banana-mcp-g3"],
    "env": {
      "GEMINI_API_KEY": "your-api-key",
      "GOOGLE_GEMINI_BASE_URL": "https://your-proxy-url.example.com"
    }
  }
}

The @google/genai SDK reads this env var and routes all API calls through the specified URL.

Development

git clone https://github.com/mascotbot/nano-banana-mcp.git
cd nano-banana-mcp
npm install
npm run dev     # Development mode
npm run build   # Production build
npm run typecheck

Tech Stack

• TypeScript, Node.js >= 20
• @google/genai SDK (^1.41.0) — Gemini 3 Pro Image
• @modelcontextprotocol/sdk — MCP protocol
• Zod — schema validation

Requirements

• Node.js 20.0.0 or higher
• Gemini API key from Google AI Studio
• Compatible with Claude Code, Cursor, and other MCP clients

License

MIT