Claude Imagine runs a Model Context Protocol (MCP) server with Nano Banana 2 as its default model — connect it once and Claude generates images inside your conversation, in Claude Code, claude.ai, or Claude Desktop.
No Google API key, no separate app, one shared credit balance.
Most guides for getting Nano Banana into Claude ask you to create your own Google or OpenRouter key first. This connector doesn't.
You sign in to Claude Imagine once through OAuth. There's no Google AI Studio project to set up, no separate billing account, and no key to paste into a config file.
The generate_image tool routes to nano-banana-2 unless you name a different model, so asking Claude for an image already uses Nano Banana 2.
Nano Banana 2 costs 3 credits per generation on free, Basic, Pro, and Studio alike — the model has no plan-tier gate. Failed generations are refunded.
The MCP tool generates from a text prompt. Image-to-image editing with Nano Banana 2 runs in the Claude Imagine web generator instead.
MCP is an open protocol for connecting an AI assistant to external tools over a live connection — this is what makes in-chat image generation possible at all.
The Claude Imagine MCP server runs at a URL (claudeimagine.com/api/mcp). Claude talks to it directly over HTTP — there's nothing to download or install on your machine.
Authentication is OAuth 2.1 with PKCE and dynamic client registration. Claude discovers the login flow automatically and opens a browser for you to sign in once.
generate_image, list_models, and get_credits — Claude calls these directly based on what you ask for, without you touching a terminal.
The connector and the web generator draw from the same balance. Buy once, spend it either place.
Pick the one you use — each takes a few minutes, once.
Run 'claude mcp add --transport http claude-imagine https://claudeimagine.com/api/mcp'. A browser opens to log in and consent, then the tools are live in your sessions.
Go to Settings → Connectors → Add custom connector, name it 'Claude Imagine', and paste the server URL from step one's pattern.
Click Connect, sign in with your Claude Imagine account, and you're set — ask Claude for an image in any chat.
Any client supporting remote HTTP transport with OAuth 2.1 (PKCE + dynamic client registration) can connect using the same server URL.
What Claude can actually call once you're connected.
Generates an image from a text prompt. Takes an optional model (nano-banana-2 default, gpt-image-2, seedream-4.5, flux-2-pro, z-image) and aspect_ratio (1:1, 4:3, 3:4, 16:9, 9:16). Failed generations are refunded automatically.
Lists available models and the live credit cost per image for your current plan tier, so Claude — and you — know the cost before generating.
Returns your remaining credit balance, so Claude can tell you if you're about to run low before starting a batch of generations.
Have another question? Email us at support@claudeimagine.com.
Connector support verified July 10, 2026 against Claude's remote MCP documentation. Tool names, models, and OAuth behavior are verified against the Claude Imagine MCP implementation.
One-time setup, then image generation lives inside every conversation.