Nano Banana MCP

A forkable MCP server that gives Claude Code, ChatGPT, and Cursor access to Google's Nano Banana image model. Generate, edit, and compose images by chatting.

Fork this template → Get a Gemini key ↗

Already forked? Finish setup in 3 steps

It takes about a minute. Until you add a Google key, the image tools will return a friendly "no key yet" message instead of an image.

Open Setup → Open the MCP tab →
  1. Add your Google (Gemini) key. In your console, open Setup → AI → Gemini and paste a key. Get one free at aistudio.google.com/apikey. It must be a Google key — an OpenAI or Anthropic key won't work here.
  2. Install the MCP endpoint. Open the MCP tab, copy the endpoint URL, and paste it into Claude / ChatGPT / Cursor's connector settings. Approve the one-time OAuth consent screen.
  3. Generate. Ask your AI: "generate a 16:9 hero image of a calm desert at dawn." Three tools light up — outputs land in your project's private storage with a preview URL.
generate_image
prompt → image

Text-to-image with Nano Banana. Aspect ratio, model variant, all configurable per call.

edit_image
image + prompt → image

Edit, restyle, recolor, or refine an existing image. Pass a URL or base64 data.

compose_images
images[] + prompt → image

Combine 2–6 source images into one composition. Style-transfer, mood-board, subject-merge.

Why this exists

Every other Nano Banana MCP wants you to self-host on Fly or Render, manage bearer tokens, redeploy on key rotations, and stitch OAuth together yourself. This template moves all of that to ~60 seconds of setup. The Gemini key never enters your code — the platform's gateway injects auth on every ai.fetch() call. The consent + OAuth flow is built into Hatchable. Outputs go to your private storage. Cost goes to your Gemini account. Revocation is one click in the MCP tab.

Built on Hatchable