Nano Banana Remote MCP Server
A remote MCP server that exposes Gemini image generation/editing as a custom connector for Claude, enabling image creation and editing from any chat interface.
README
Nano Banana — Remote MCP Server
A small remote MCP server that exposes Gemini image generation/editing as a custom connector you can add to Claude on web, mobile, and desktop — so it works in every chat, not just on one PC.
Tools:
generate_image(prompt)— returns a new image inline in the chat.edit_image(imageUrl, prompt)— edits an image from a public URL, returns it inline.
How it differs from the local setup
The local nano-banana server runs node on your PC and only works in Claude
Desktop / Claude Code on that machine. This remote version runs on an always-on
host at a public URL, which is what claude.ai (web + mobile) requires.
Environment variables
| Var | Required | Purpose |
|---|---|---|
GEMINI_API_KEY |
yes | Your Google Gemini API key (kept server-side) |
MCP_AUTH_TOKEN |
recommended | Shared secret. If set, requests must include it (see below) |
GEMINI_IMAGE_MODEL |
no | Defaults to gemini-2.5-flash-image |
PORT |
no | Host sets this automatically |
How auth works
If MCP_AUTH_TOKEN is set, the token must be supplied one of three ways. For
the claude.ai connector dialog (URL only), put it in the path:
https://YOUR-APP-URL/mcp/YOUR_TOKEN
(Also accepted: ?t=YOUR_TOKEN, or header Authorization: Bearer YOUR_TOKEN.)
Deploy — Render (free, recommended for simplicity)
- Put this
remote-server/folder in a GitHub repo. - Go to https://render.com → New → Web Service → connect the repo.
- Render auto-detects
render.yaml. Settings if asked: Runtime Node, Buildnpm install, Startnpm start. - Under Environment, add secrets:
GEMINI_API_KEY= your keyMCP_AUTH_TOKEN= a long random string (your choice)
- Deploy. You get a URL like
https://nano-banana-remote-mcp.onrender.com. - Your connector URL is:
https://nano-banana-remote-mcp.onrender.com/mcp/YOUR_TOKEN
Render's free tier sleeps after ~15 min idle, so the first call after a nap takes ~30–50s (it may need a retry). The $7/mo Starter tier stays always-on. Fly.io and Railway are alternatives if you prefer.
Verify it's live
Open https://YOUR-APP-URL/health in a browser — you should see
{"ok":true,"model":"gemini-2.5-flash-image"}.
Add it to Claude (works everywhere after this)
- Go to claude.ai → Settings → Connectors (Pro/Max/Team/Enterprise plan required for custom connectors).
- Add custom connector.
- Name:
Nano Banana. URL:https://YOUR-APP-URL/mcp/YOUR_TOKEN. - Save and enable it. It now syncs to web, mobile, and desktop.
- In any chat: "generate an image of a banana on a surfboard".
Local test
cp .env.example .env # fill in GEMINI_API_KEY, MCP_AUTH_TOKEN
npm install
npm start
node test-client.mjs # exercises the MCP protocol against localhost
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。