# nanobanana-mcp [Health: Active]

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/getjoystick/nanobanana-mcp  
**GitHub Stars:** 1  
**npm Downloads (last month):** 188  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/nanobanana-mcp

## Description
Lean MCP server for Google's Nano Banana (Gemini) image models. Returns file paths, not base64.

## Tools
Capabilities this server exposes over MCP:

- **generate_image**
- **edit_image** — Same params, plus:
- **list_models** — Takes no params. Lists the image and video model ids your API key can use, in two

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "nanobanana-mcp": {
    "command": "npx",
    "args": ["-y","@nanobanana/mcp"],
    "env": {
      "GEMINI_API_KEY": "",
      "GOOGLE_API_KEY": "",
      "NANOBANANA_MODEL": "",
      "NANOBANANA_VIDEO_MODEL": "",
      "NANOBANANA_OUTPUT_DIR": ""
    }
  }
}
```

**Requires environment variables:** `GEMINI_API_KEY`, `GOOGLE_API_KEY`, `NANOBANANA_MODEL`, `NANOBANANA_VIDEO_MODEL`, `NANOBANANA_OUTPUT_DIR` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What nanobanana-mcp does

The nanobanana-mcp MCP server exposes Google's Nano Banana image models and Veo video models to MCP-compatible clients. It can generate images from prompts, edit existing images with one or more reference files, create video jobs, extend generated Veo clips, check video job status, and list model IDs available to the configured API key.

Image operations save results to disk and return file paths rather than embedding base64 data in the tool response. The image tools support prompt-based creation, selectable aspect ratios, output sizes, optional Google Search grounding, and model selection. Editing accepts PNG, JPG, WebP, HEIC, or PDF inputs; the first image is the base input and additional files can act as references.

## How it works

The nanobanana-mcp MCP server runs as a local stdio process, commonly started with `npx -y @nanobanana/mcp`. An MCP client sends tool calls, while the server uses the Gemini API key from the process environment. Generated images and completed videos are written to an output directory, which defaults to `./nano-banana` and can be changed through configuration.

Video creation is asynchronous because Veo jobs can take one to six minutes. `generate_video` or `extend_video` returns a job ID; `get_video_result` waits for up to 45 seconds per call and reports whether processing is complete. A completed result includes a local MP4 path and a Veo URI. The URI is required for later extensions and expires with Google's server-side copy, generally after about two days.

## Setup and configuration

Create a Gemini API key and provide it as `GEMINI_API_KEY`; `GOOGLE_API_KEY` is also supported. Claude Code can register the server with an environment variable and the `npx` command. Claude Desktop and Cursor use an MCP configuration entry with `npx` as the command and the API key in the environment section.

The default image model is `gemini-3.1-flash-image`, and the default video model is `veo-3.1-fast-generate-preview`. `NANOBANANA_MODEL`, `NANOBANANA_VIDEO_MODEL`, and `NANOBANANA_OUTPUT_DIR` override those defaults. The server also avoids writing to the filesystem root when a desktop client starts it from `/`.

## Tools and capabilities

The nanobanana-mcp MCP server includes six tools:

- `generate_image` creates an image from a prompt.
- `edit_image` modifies an input image and can use multiple references.
- `generate_video` starts a text-to-video or image-to-video job with optional audio, references, and negative prompts.
- `extend_video` continues a Veo clip by about seven seconds using its Veo URI.
- `get_video_result` polls a job and saves a completed MP4.
- `list_models` reports image and video model IDs usable by the API key.

Image output sizes include 512, 1K, 2K, and 4K. Image aspect ratios range from portrait and square formats to wide formats, while Veo supports 16:9 and 9:16. Veo clips can be four, six, or eight seconds. Higher video resolutions require an eight-second duration, and reference images require eight seconds. Veo Lite does not support extensions or reference images.

## Limitations and notes

Veo has no free tier and requires a billing-enabled API key. Video generation, extension, reference-image support, and resolution options vary by Veo model. Image-based video inputs must be PNG or JPG, and an input image is cropped to the selected video aspect ratio.

Download completed videos promptly because Google's server-side copies are temporary. Keep the returned Veo URI if the clip may be extended. A pinned model ID can become unavailable; use `list_models` to inspect the IDs currently allowed for the API key. The project is licensed under MIT.

_Full upstream README: https://allmcps.com/mcp/nanobanana-mcp/readme_