MCP Server - Vercel AI SDK

The Vercel AI SDK exposes MCP tools through its provider-independent tool interface. Install AI SDK Core, the Anthropic provider, and the MCP client:


pnpm add ai @ai-sdk/anthropic @ai-sdk/mcp

Create the MCP client, discover the Xweather tools, and pass them to generateText:


// app/api/weather/route.ts
import { anthropic } from "@ai-sdk/anthropic";
import { createMCPClient } from "@ai-sdk/mcp";
import { generateText, isStepCount } from "ai";
 
export async function POST(req: Request) {
    const { location } = await req.json();
 
    const mcpClient = await createMCPClient({
        transport: {
            type: "http",
            url: "https://mcp.api.xweather.com/mcp?include_tools=xweather_get_current_weather,xweather_get_weather_conditions",
            headers: {
                Authorization: `Bearer ${process.env.XWEATHER_API_KEY}`,
            },
        },
    });
 
    try {
        const tools = await mcpClient.tools();
 
        const { text } = await generateText({
            model: anthropic("claude-sonnet-4-6"),
            tools,
            stopWhen: isStepCount(5),
            prompt: `Give me the next 6 hours of weather for ${location}`,
            onToolExecutionStart({ toolCall }) {
                console.info("Calling Xweather MCP tool", {
                    name: toolCall.toolName,
                    input: toolCall.input,
                });
            },
        });
 
        return Response.json({ answer: text });
    } finally {
        await mcpClient.close();
    }
}

Best practices

Restrict the tools exposed to the model with Xweather’s include_tools or include_tags query parameters.
Close short-lived MCP clients in a finally block. For streamed responses, close the client in the onEnd callback.
Store XWEATHER_API_KEY in Vercel project environment variables; never expose it to the browser.
Use onToolExecutionStart and onToolExecutionEnd to capture tool arguments, results, timing, and errors for auditing.