LabscoMCP early access ↗
petrkindlmannMCP SERVER

Nano Banana MCP

Gemini's image models with multi-turn editing — generate, then keep editing the same image conversationally.

by petrkindlmann ↗Image GenerationCommunity

LABSCO SUMMARY

Iteration is the feature, not generation.

Anything can generate an image from a prompt; what makes this usable is that the second request keeps the first one's context, so "same scene, different light" actually stays the same scene. The tier table is worth reading before you commit — flash is the only one that takes video, pro is the one that renders text you can read, and nano exists for volume.

What it is

An image generation and editing server over Google's Gemini image models. Each result comes back with an interaction_id, so the next instruction — heavier fog, add a deer — continues the same conversation rather than starting a new image from scratch.

What you get

  • generate_image — a prompt, an output path, and control over ratio, size up to 4K, and grounding
  • edit_image — chain from previous_interaction_id, or pass reference_images from disk, up to 14 inputs
  • generate_story — one prompt becomes a numbered sequence of captioned images: storyboards, comics, recipes
  • generate_icon_set — one prompt per icon, chained so the whole set shares a look
  • generate_from_video — a poster or thumbnail from a public YouTube URL
  • use_search grounds an image in current data through Google Search; use_image_search adds visual context
  • Three tiers with different capabilities — nano at 1K and PNG only, flash for the widest ratios and video input, pro for the highest quality and legible text inside the image
  • Inline previews returned to the client by default, so the model can see what it produced and correct itself

Requirements

Node.js 18+ and a Gemini API key from Google AI Studio, set as GEMINI_API_KEY. Clone, npm install, then register index.js with your client. One thing to check first: the Interactions API is in beta and the Gemini 3 tiers may need access on your key — set model: "nano" if flash or pro come back unavailable. Model IDs can be repointed per tier with NANO_BANANA_MODEL_FLASH and its siblings when Google rotates them. Version 2.1.0, MIT licensed.

Setup effort

Build from source — clone the repository, then follow the project's own run instructions

See installation options