> ## Documentation Index
> Fetch the complete documentation index at: https://docs.corsair.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Vercel AI SDK

> Connect Corsair to the Vercel AI SDK over HTTP.

Use `createVercelAiMcpClient` to connect Corsair to the [Vercel AI SDK](https://sdk.vercel.ai) via HTTP transport.

Unlike the direct SDK adapters, Vercel AI connects over HTTP — you expose Corsair as an MCP server endpoint and the client connects to it.

## Install

```bash theme={null}
npm install ai
```

## Server

Expose Corsair as an MCP HTTP endpoint using `createBaseMcpServer` and `createMcpRouter`.

```ts server.ts theme={null}
import express from 'express';
import { createBaseMcpServer, createMcpRouter } from '@corsair-dev/mcp';
import { corsair } from './corsair';

const app = express();
app.use(express.json());

app.use('/mcp', createMcpRouter(() => createBaseMcpServer({ corsair })));

app.listen(3000, () => console.log('MCP server running on :3000'));
```

## Client

Connect from your Vercel AI application using `createVercelAiMcpClient`.

```ts agent.ts theme={null}
import { generateText, stepCountIs } from 'ai';
import { anthropic } from '@ai-sdk/anthropic';
import { createVercelAiMcpClient } from '@corsair-dev/mcp';

const client = await createVercelAiMcpClient({
    url: 'http://localhost:3000/mcp',
});

const tools = await client.tools();

const { text } = await generateText({
    model: anthropic('claude-sonnet-4-6'),
    tools,
    prompt: 'Setup corsair, then list all Slack channels.',
    stopWhen: stepCountIs(10),
});

console.log(text);
await client.close();
```

`createVercelAiMcpClient` returns a client that speaks the MCP protocol over HTTP. Call `client.tools()` to retrieve the tool definitions, then pass them to any Vercel AI `generateText` or `streamText` call.

## AI SDK 6

In AI SDK 6, `maxSteps` was replaced by `stopWhen`. Without it, tool-calling agents may never complete or return output.

Use `stopWhen: stepCountIs(10)` to cap the number of tool-call steps, or `stopWhen: isLoopFinished()` to run until the model finishes naturally:

```ts theme={null}
import { generateText, isLoopFinished } from 'ai';

const { text } = await generateText({
  model: anthropic('claude-sonnet-4-6'),
  tools,
  prompt: 'List my GitHub repos with the most open issues.',
  stopWhen: isLoopFinished(),
});
```