MCP Boilerplate Next.js
A minimal Next.js starter for building MCP Apps, enabling MCP hosts to render interactive UIs alongside tool calls via an MCP server endpoint and a React widget UI.
README
MCP Apps — Minimal Next.js Starter
A minimal Next.js starter for building MCP Apps — interactive UIs that MCP hosts render alongside tool calls.
How it works
The Next.js app serves two roles:
- MCP server (
app/mcp/route.ts) — registers tools and resources viamcp-handlerand@modelcontextprotocol/ext-apps. - Widget UI (
app/page.tsx) — a React page that MCP hosts render inside a sandboxed iframe. The MCP route self-fetches the rendered page HTML and serves it as an MCP resource.
The useMcpApp hook (app/hooks/use-mcp-app.ts) connects to the host via the App class from @modelcontextprotocol/ext-apps and provides tool input/result data as React state.
Getting started
pnpm install
Local development with a tunnel
MCP Apps need a public HTTPS URL. Use ngrok (or any tunnel):
ngrok http 3000
Copy the HTTPS URL and set it in .env:
BASE_URL=https://xxxx-xxx-xxx.ngrok-free.app
Then start the dev server:
pnpm dev
Connect to a host
Add your MCP server URL to any MCP host that supports Apps:
https://xxxx-xxx-xxx.ngrok-free.app/mcp
For example, in ChatGPT, Cursor, or Claude.ai: Settings > Apps > add the URL above.
Project structure
app/
page.tsx — Homepage (widget UI)
about/page.tsx — Example sub-page (navigation demo)
counter/page.tsx — Example sub-page (interactivity demo)
mcp/route.ts — MCP server endpoint
hooks/use-mcp-app.ts — React hook for the MCP Apps bridge
layout.tsx — Root layout with iframe bootstrap patches
baseUrl.ts — Public URL resolver (tunnel / Vercel)
middleware.ts — CORS headers for cross-origin iframe access
next.config.ts — assetPrefix for iframe asset loading
Key files
app/mcp/route.ts— Define your tools and resources here. Thegreettool is a minimal example.app/page.tsx— Your widget UI. Edit this like any Next.js page. It receives tool data viauseMcpApp().app/hooks/use-mcp-app.ts— SingletonAppbridge withsessionStoragepersistence. ProvidestoolInput,toolResult, andconnectedstate.
Deploy
Deploy to Vercel — no additional configuration needed. The BASE_URL is automatically derived from Vercel's environment variables in production.
Once deployed, connect it to any MCP host using:
https://your-app.vercel.app/mcp
For example, in ChatGPT, Cursor, or Claude.ai: Settings > Apps > add the URL above as a connector.
Learn more
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。