A Vercel adapter for the Model Context Protocol (MCP), enabling real-time communication between your applications and AI models. Currently supports Next.js with more framework adapters coming soon.
npm install @vercel/mcp-adapter @modelcontextprotocol/sdk # or yarn add @vercel/mcp-adapter @modelcontextprotocol/sdk # or pnpm add @vercel/mcp-adapter @modelcontextprotocol/sdk # or bun add @vercel/mcp-adapter @modelcontextprotocol/sdk
// app/api/[transport]/route.ts import { createMcpHandler } from "@vercel/mcp-adapter"; const handler = createMcpHandler( (server) => { server.tool( "roll_dice", "Rolls an N-sided die", { sides: z.number().int().min(2), }, async ({ sides }) => { const value = 1 + Math.floor(Math.random() * sides); return { content: [{ type: "text", text: `🎲 You rolled a ${value}!` }], }; } ); }, { // Optional server options }, { // Optional redis config redisUrl: process.env.REDIS_URL, basePath: "/api", // this needs to match where the [transport] is located. maxDuration: 60, verboseLogs: true, } ); export { handler as GET, handler as POST };
Depending on the version of your client application, remote MCP's may need to use mcp-remote to proxy Streamable HTTP into stdio.
If your client supports it, it's recommended to connect to the Streamable HTTP endpoint directly such as:
"remote-example": { "url": "http://localhost:3000/api/mcp", }
Due to client versions, and varying levels of support, you can list mcp-remote
as the method for end users to connect to your MCP server.
The above set up snippet will then look like:
"remote-example": { "command": "npx", "args": [ "-y", "mcp-remote", "http://localhost:3000/api/mcp" // this is your app/api/[transport]/route.ts ] }
When you want to use it in your MCP client of choice:
Depending on the version of your client application, remote MCP's may need to use mcp-remote to proxy Streamable HTTP into Stdio.
In order to add an MCP server to Claude Desktop you need to edit the configuration file located at:
"remote-example": { "command": "npx", "args": [ "-y", "mcp-remote", "http://localhost:3000/api/mcp" // this is your app/api/[transport]/route.ts ] }
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json If it does not exist yet, you may need to enable it under Settings > Developer.
Restart Claude Desktop to pick up the changes in the configuration file. Upon restarting, you should see a hammer icon in the bottom right corner of the input box.
The configuration file is located at ~/.cursor/mcp.json.
As of version 0.48.0, Cursor supports unauthed SSE servers directly. If your MCP server is using the official MCP OAuth authorization protocol, you still need to add a "command" server and call mcp-remote.
The configuration file is located at ~/.codeium/windsurf/mcp_config.json.
- Use the MCP client in your application:
// app/components/YourComponent.tsx import { McpClient } from "@modelcontextprotocol/sdk/client"; const client = new McpClient({ // When using basePath, the SSE endpoint will be automatically derived transport: new SSEClientTransport("/api/mcp/mcp"), }); // Use the client to make requests const result = await client.request("yourMethod", { param: "value" });
The initializeMcpApiHandler
function accepts the following configuration options:
interface Config { redisUrl?: string; // Redis connection URL for pub/sub basePath?: string; // Base path for MCP endpoints maxDuration?: number; // Maximum duration for SSE connections in seconds verboseLogs?: boolean; // Log debugging information }
The MCP adapter supports the MCP Authorization Specification per the through the withMcpAuth
wrapper. This allows you to protect your MCP endpoints and access authentication information in your tools.
// app/api/[transport]/route.ts import { createMcpHandler, withMcpAuth } from "@vercel/mcp-adapter"; // Create your handler as normal const handler = createMcpHandler( (server) => { server.tool( "echo", "Echo a message", { message: z.string() }, async ({ message }, extra) => { // Access auth info in your tools via extra.authInfo return { content: [ { type: "text", text: `Echo: ${message}${ extra.authInfo?.token ? ` for user ${extra.authInfo.clientId}` : "" }`, }, ], }; } ); }, { // Optional server options } ); // Wrap your handler with authorization const verifyToken = async ( req: Request, bearerToken?: string ): Promise<AuthInfo | undefined> => { if (!bearerToken) return undefined; // Replace this example with actual token verification logic // Return an AuthInfo object if verification succeeds // Otherwise, return undefined const isValid = bearerToken.startsWith("__TEST_VALUE__"); if (!isValid) return undefined; return { token: bearerToken, scopes: ["read:stuff"], // Add relevant scopes clientId: "user123", // Add user/client identifier extra: { // Optional extra information userId: "123", }, }; }; // Make authorization required const authHandler = withMcpAuth(handler, verifyToken, { required: true, // Make auth required for all requests requiredScopes: ["read:stuff"], // Optional: Require specific scopes resourceMetadataPath: "/.well-known/oauth-protected-resource", // Optional: Custom metadata path }); export { authHandler as GET, authHandler as POST };
When implementing authorization in MCP, you must define the OAuth Protected Resource Metadata endpoint. This endpoint provides information about your MCP server's authentication requirements to clients.
Create a new file at app/.well-known/oauth-protected-resource/route.ts
:
import { protectedResourceHandler, metadataCorsOptionsRequestHandler, } from "@vercel/mcp-adapter"; const handler = protectedResourceHandler({ // Specify the Issuer URL of the associated Authorization Server authServerUrls: ["https://auth-server.com"], }); export { handler as GET, metadataCorsOptionsRequestHandler as OPTIONS };
This endpoint provides:
resource
: The URL of your MCP serverauthorization_servers
: Array of OAuth authorization server Issuer URLs that can issue valid tokens
The path to this endpoint should match the resourceMetadataPath
option in your withMcpAuth
configuration, which by default is /.well-known/oauth-protected-resource
(the full URL will be https://your-domain.com/.well-known/oauth-protected-resource
).
- Client makes a request with a Bearer token in the Authorization header
- The
verifyToken
function validates the token and returns auth info - If authentication is required and fails, a 401 response is returned
- If specific scopes are required and missing, a 403 response is returned
- On successful authentication, the auth info is available in tool handlers via
extra.authInfo
- Framework Support: Currently supports Next.js with more framework adapters coming soon
- Multiple Transport Options: Supports both Streamable HTTP and Server-Sent Events (SSE) transports
- Redis Integration: For SSE transport resumability
- TypeScript Support: Full TypeScript support with type definitions
- Next.js 13 or later (for Next.js adapter)
- Node.js 18 or later
- Redis (optional, for SSE transport)
Apache-2.0