Banana Art Lab MCP Server
Generate images and build canvases from Codex, Claude Code, Cursor or any MCP client, with GPT Image 2, Nanobanana Pro, Seedream and the other site models, paid with your Banana Art Lab credits.
https://trynanobanana.net/mcpConnect your assistant
Codex
Add the server, then sign in. Codex opens trynanobanana.net in your browser; choose Allow.
codex mcp add banana-art-lab --url https://trynanobanana.net/mcpcodex mcp login banana-art-labClaude Code
Add the server, then type /mcp in Claude Code, pick banana-art-lab and sign in.
claude mcp add --transport http banana-art-lab https://trynanobanana.net/mcpCursor
Add it to ~/.cursor/mcp.json (or the project's .cursor/mcp.json). Cursor asks you to sign in the first time.
{
"mcpServers": {
"banana-art-lab": {
"url": "https://trynanobanana.net/mcp"
}
}
}Other clients, or an API key
Add https://trynanobanana.net/mcp as a remote HTTP MCP server. Clients that support OAuth ask you to sign in. For the others, create an API key in the canvas under Settings → AI tools and send it as a bearer token. In Codex that looks like this, with the key in the BANANA_ART_LAB_API_KEY environment variable:
[mcp_servers.banana-art-lab]
url = "https://trynanobanana.net/mcp"
bearer_token_env_var = "BANANA_ART_LAB_API_KEY"What your assistant can do
| Tool | What it does |
|---|---|
| generate_image | Generates 1-4 images from a prompt and optional reference images, with any site model, aspect ratio and resolution. Optionally places the prompt and results on a canvas. |
| get_generation | Fetches the images of a generation that took longer than about 45 seconds. |
| list_models | Lists the models, resolutions, quality levels and credits per image. |
| get_account | Shows the credit balance of the connected account. |
| list_canvases | Lists your canvases with their ids. |
| get_canvas | Reads a canvas: its nodes (text, images, positions) and connections. |
| create_canvas | Creates a new canvas. |
| edit_canvas | Adds, updates, connects and deletes nodes on a canvas. |
Things to ask
- “Generate four product shots of a ceramic coffee mug on white marble with GPT Image 2 Saver, and put them on a new canvas called "Mug launch".”
- “Read my canvas "Spring campaign" and make a 16:9 version of every image on it.”
- “List the Banana Art Lab models with their prices, then make a 2K poster of a night market with Nanobanana Pro.”
- “Turn this product description into three prompt nodes on my canvas, connect each to an image, and generate them.”
Your account stays in your hands
- You sign in and approve each app on trynanobanana.net. Apps never see your password.
- Access tokens last an hour and renew themselves; disconnect an app or revoke a key at any time in Canvas Settings → AI tools.
- The server only generates images when your assistant asks, at the same credit prices as the website (see models and pricing).
- It works with your Banana Art Lab account only; it cannot reach files on your computer.
Questions
What is the Banana Art Lab MCP server?
MCP (Model Context Protocol) is how AI assistants such as Codex, Claude Code and Cursor use outside tools. The Banana Art Lab MCP server at https://trynanobanana.net/mcp gives them image generation and your canvases, so you can ask for images in the same chat where you work.
Which apps can connect?
Any MCP client that supports remote (Streamable HTTP) servers. Codex, Claude Code and Cursor sign in with OAuth: you approve the app on trynanobanana.net once. Clients without OAuth can use an API key from Canvas Settings → AI tools.
What does it cost?
The tools are free to connect. Each generated image uses credits at the same price as on the website, for example 2 credits for GPT Image 2 Saver at any resolution. New accounts get 10 free credits when they sign up, and you can buy more on the pricing page.
Can the AI see my canvases and spend my credits?
Only after you allow it on the consent page, and only on the account you approved. It cannot reach files on your computer. You can disconnect an app, or revoke an API key, at any time in Canvas Settings → AI tools.
Why does a generation sometimes come back as a job?
Codex waits about 60 seconds for a tool by default. If images are not ready within about 45 seconds, generate_image returns a job id and your assistant fetches the images with get_generation a little later. Nothing is charged twice.
Where do canvas changes show up?
In your canvas at trynanobanana.net/canvas. If it is open, changes from your assistant appear within about ten seconds; otherwise they are there the next time you open it.