MCP is an open protocol that lets AI assistants call external tools. Connect Nano Banana to an MCP-capable assistant, describe what you need in plain words, and the assistant can look up models, submit a generation, track the task and fetch the image without you switching between tabs.
Ask for cover art, product shots and illustrations in the same conversation where you write, plan or build a web page.
The assistant checks the current image models and parameters, submits the task and polls progress, so you can focus on the content and the result.
Uses your existing credits with the same per-image pricing as the website and OpenAPI, and failed generations are refunded. Connecting MCP never starts a generation by itself.
Comparing with Higgsfield? See our Higgsfield alternative comparison for prices, models and MCP sign-in side by side.
Pick your client and press the button. We create a dedicated API key, put it into a setup prompt and copy it; send it to your assistant and it configures Nano Banana MCP for you.
Each press creates a new dedicated key. Above 10 keys the oldest one-click key is revoked automatically (keys you created by hand are never touched); manage them on the API page. The prompt contains your key, so only send it to your own AI tools.
You need a Nano Banana account that has made a purchase, a valid API key, and an MCP client that supports remote Streamable HTTP with custom request headers.
API and MCP access are available to paid accounts only; bonus credits alone cannot be used. Create a dedicated API key on the OpenAPI page and put it only in a trusted client's connection settings, never in a chat.
In your AI tool's MCP settings, add a remote HTTP server with the URL and Authorization header below. Do not choose stdio mode, which only runs local commands.
After connecting, ask the assistant to list the available models first. Once the three tools below appear, describe the image you want and explicitly allow it to spend credits.
Replace YOUR_NANO_BANANA_API_KEY with your key. This page does not collect or store keys.
{
"mcpServers": {
"nano-banana": {
"url": "https://www.nano-banana.cn/api/mcp/",
"headers": {
"Authorization": "Bearer YOUR_NANO_BANANA_API_KEY"
}
}
}
}The JSON example works for clients that accept url / headers; Claude Code uses a single command; the Codex example reads the key from an environment variable. Exact field names and config file locations follow each client's documentation.
MCP support does not guarantee support for this server's authentication. Make sure your client offers remote HTTP and a custom Authorization header.
Never commit a config file that contains a real key to a code repository.
Official setup guideThe command stores the key in your local Claude Code config, so only run it on your own computer.
Official setup guideRemote or cloud sessions need this variable set in the environment that actually runs Codex.
Official setup guideThe JSON above is a reference for the connection fields, not a one-click import verified for every WorkBuddy version.
A ChatGPT subscription does not include Nano Banana generation credits.
Official setup guideState the purpose, size and reference images first, then confirm the model and cost. These examples are prompts only; copying them does not start a generation.
First list the image models available on Nano Banana, pick one that suits an article cover and tell me how many credits each image costs. Theme: "slow living in the city", landscape 16:9, natural light, no text. After I confirm, generate just 1 image and return the task ID and the image URL.Here is a public link to my product photo: https://example.com/product.png . Check which Nano Banana models support image-to-image and their prices. After I confirm, generate 1 clean e-commerce shot on a white background that keeps the product's look.I need images for a 5-section tutorial. First list a scene description and a recommended model for each section plus the total credits. After I confirm, submit them one by one, submit each only once, and query the same task ID until each image is ready.Nano Banana MCP currently offers image generation only. Reference images must be publicly accessible links; whether local files can be read or results saved depends on your AI client's own file and network permissions.
Connecting and listing models never submit a generation. Only generate_media spends credits, based on the chosen model and parameters, at the same price as the website and OpenAPI. Failures, timeouts and moderation rejections are refunded. Your AI tool's own subscription is billed separately.
Like the OpenAPI, it is open to paid accounts only: any past top-up, subscription or renewal counts. Accounts with bonus credits only receive a 403 message. No captcha is required when calling with an API key.
Make sure you use https://www.nano-banana.cn/api/mcp/, remote HTTP mode and a valid API key. The header must read Authorization: Bearer, a space, then the key. Never put the key in the URL. If your client cannot send custom headers or only supports OAuth, it cannot connect with this setup.
A network timeout does not mean the generation failed. Have the assistant keep the data.id returned by generate_media and query that same task at least 10 seconds apart instead of resubmitting. If you cannot tell whether a submission went through, check your task history on the website.
Not yet. Like the OpenAPI, Nano Banana MCP exposes image models only. For reference images, provide public http(s) image links; the server copies them before generating.
Use a separate key for each tool and store it only in that tool's secure config. Never paste it into chats, public repositories, screenshots or logs. Send the key only to nano-banana.cn and never when downloading images. If it leaks, disable it in the console right away and create a new one.
Create a dedicated API key, connect a supported client and start by listing models.