Nano Banana MCP
by brandon695
README.md
# Nano Banana MCP (Local Fork)
Fork of [nano-banana-mcp](https://github.com/ConechoAI/Nano-Banana-MCP) with model selection and aspect ratio support.
## Changes from upstream
- **Model selection**: Choose between `flash` (fast/cheap) and `pro` (slower/better) via env var or per-request parameter
- **Aspect ratio**: Pass `aspectRatio` to `generate_image` (1:1, 16:9, 9:16, 3:2, 4:3, 21:9, etc.)
- **Accurate model labeling**: Status output shows actual model name instead of incorrect "Gemini 2.5 Flash"
## Models
| Quality | Model ID | Speed | Cost |
|---------|----------|-------|------|
| `flash` | `gemini-3.1-flash-image-preview` | Fast | ~$0.04-0.05/image |
| `pro` | `gemini-3-pro-image-preview` | Slower | ~$0.13-0.24/image |
## Configuration
```json
{
"nano-banana": {
"type": "stdio",
"command": "node",
"args": ["/path/to/nano-banana-mcp-local/dist/index.js"],
"env": {
"GEMINI_API_KEY": "your-key-here",
"NANO_BANANA_MODEL": "flash"
}
}
}
```
### Environment Variables
| Variable | Values | Default | Description |
|----------|--------|---------|-------------|
| `GEMINI_API_KEY` | API key string | required | Your Google AI Studio API key |
| `NANO_BANANA_MODEL` | `flash`, `pro` | `flash` | Default model for all requests |
## Usage
### generate_image
```
generate_image({
prompt: "A dramatic photo of...",
quality: "pro", // optional, overrides env default
aspectRatio: "16:9" // optional, default 1:1
})
```
### edit_image / continue_editing
Same as upstream, plus optional `quality` parameter.
## License
MIT (same as upstream)