# apifare for Vercel AI SDK

One prepaid, capped key for every API your agent needs, under hard caps and a kill switch.

1. Get a token: https://apifare.com/start?origin=page:integrations-vercel-ai-sdk (shown once).
2. Connect with createMCPClient from @ai-sdk/mcp (not the older experimental_createMCPClient from ai) and pass its tools to generateText. The token is read from the APIPAY_TOKEN environment variable and sent as an Authorization header.

```
npm i @ai-sdk/mcp ai zod
```

```
import { createMCPClient } from '@ai-sdk/mcp';
import { generateText, isStepCount } from 'ai';

const mcpClient = await createMCPClient({
  transport: {
    type: 'http',
    url: 'https://apifare.com/mcp',
    headers: { Authorization: `Bearer ${process.env.APIPAY_TOKEN}` },
  },
});

try {
  const tools = await mcpClient.tools();
  const { text } = await generateText({
    model: 'openai/gpt-5.4',
    tools,
    stopWhen: isStepCount(10),
    prompt: "What's the weather in London?",
  });
  console.log(text);
} finally {
  await mcpClient.close();
}
```

   Set APIPAY_TOKEN, then run it as an ES module (for the top-level await). The finally block closes the client, as the AI SDK docs ask.
3. Try it: "search the web for X", then "fetch a URL", then "what's the weather in London".
4. Caps and the kill switch are checked before dispatch. An empty balance returns HTTP 402 with a top-up link (see /errors.md).

Prices: https://apifare.com/capabilities. Config checked against https://ai-sdk.dev/docs/ai-sdk-core/mcp-tools on 2026-09-13.