mcp-nextjs
Next.js-based MCP server with OAuth 2.1 authentication, deployable on Vercel, supporting SSE and Streamable HTTP transports for various clients.
README
OAuth 2.1 MCP Server as a Next.js app on Vercel
This is a Next.js-based application that provides an MCP (Model Context Protocol) server with OAuth 2.1 authentication support. It is intended as a model for building your own MCP server in a Next.js context. It uses the @vercel/mcp-adapter to handle the MCP protocol, in order to support both SSE and Streamable HTTP transports.
In addition to being an OAuth server, it also requires the user authenticate. This is currently configured to use Google as a provider, but you could authenticate users however you want (X, GitHub, your own user/password database etc.) without breaking the OAuth flow.
Using with
Claude Desktop and Claude.ai
Claude currently supports only the older SSE transport, so you need to give it a different URL to all the other clients listed here.
Use the "Connect Apps" button and select "Add Integration". Provide the URL like https://example.com/mcp/sse (the /sse at the end is important!). Note that Claude Desktop and Web will not accept a localhost URL.
Cursor
Edit your mcp.json to look like this:
{
"mcpServers": {
"MyServer": {
"name": "LlamaIndex MCP Demo",
"url": "https://example.com/mcp/mcp",
"transport": "http-stream"
},
}
}
VSCode
VSCode currently doesn't properly evict the client ID, so client registration fails if you accidentally delete the client (the workaround in that issue will resolve it). Otherwise, it works fine. Add this to your settings.json:
"mcp": {
"servers": {
"My Server": {
"url": "https://example.com/mcp/mcp"
}
}
}
MCP Inspector
Tell Inspector to connect to https://example.com/mcp/mcp, with Streamable HTTP transport. You can also use the SSE transport by connecting to https://example.com/mcp/sse instead.
Running the server
npm install
prisma generate
npm run dev
The very first time you will also need to run prisma db push to create the database tables.
Environment variables
Required environment variables should be in .env: (not .env.local because Prisma doesn't support it)
DATABASE_URL="postgresql://user:pass@server/database"
AUTH_SECRET="any random string"
GOOGLE_CLIENT_ID="a Google OAuth client ID"
GOOGLE_CLIENT_SECRET="a Google OAuth client secret"
REDIS_URL="rediss://user:pass@host:6379"
DATABASE_URL is required for OAuth authentication to work, this is where sessions etc. live.
REDIS_URL is required if you need SSE transport to work (i.e. you want to support Claude Desktop and Web).
Architecture
If you're using this as a template for your own Next.js app, the important parts are:
/src/app/api/oauth/*- these implement oauth client registration and token exchange/src/app/oauth/authorize/page.tsx- this implements the oauth consent screen (it's extremely basic right now)/src/mcp/[transport]/route.ts- this implements the MCP server itself. Your tools, resources, etc. should be defined here.
To handle OAuth your app needs to be able to persist clients, access tokens, etc.. To do this it's using a PostgreSQL database accessed via Prisma. You can swap this for some other database if you want (it will be easiest if it's another Prisma-supported database).
You'll also notice:
src/app/auth.ts- this implements Auth.js authentication to your app itself. It's configured to use Google as a provider, but you can change it to use any other provider supports by Auth.js. This is not required for the MCP server to work, but it's a good idea to have it in place for your own app.src/app/api/auth/[...nextauth]/route.ts- this plumbs in the Auth.js authentication, and is again not part of the OAuth implementation.
Deploying to production
This app only works if deployed to Vercel currently, due to its dependence on the @vercel/mcp-adapter package, which in turn is required to support the old SSE transport. We didn't feel like implementing a whole extra protocol just for Claude Desktop.
Deploy as usual. You'll need to add prisma generate to your build command, and of course you'll need all the same environment variables as in the development environment.
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。