Skip to main content

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.

Use createVercelAiMcpClient from @corsair-dev/app when your app uses the Vercel AI SDK. The helper connects to Corsair’s hosted HTTP MCP endpoint and returns tools in the shape expected by streamText and generateText.

​
Install

npm install @corsair-dev/app ai @ai-sdk/mcp @ai-sdk/openai

​
Example

import "dotenv/config";

import { openai } from "@ai-sdk/openai";
import { createClient, createVercelAiMcpClient } from "@corsair-dev/app";
import { stepCountIs, streamText } from "ai";

const corsair = createClient({
  apiKey: process.env.CORSAIR_API_KEY!,
});

const tenant = corsair
  .instance(process.env.CORSAIR_INSTANCE_ID!)
  .tenant("dev");

const conn = await tenant.mcpKeys.connection();

const mcpClient = await createVercelAiMcpClient({
  url: conn.mcpHttpUrl,
  apiKey: process.env.CORSAIR_MCP_API_KEY!,
});

const tools = await mcpClient.tools();

const stream = streamText({
  model: openai("gpt-4.1"),
  prompt: "Star the corsairdev/corsair repo on GitHub",
  stopWhen: stepCountIs(10),
  tools,
});

for await (const textPart of stream.textStream) {
  process.stdout.write(textPart);
}

await mcpClient.close?.();

​
Common mistakes

mcpClient.tools() is async. Always await it before passing tools to streamText: 
const tools = await mcpClient.tools();
Do not do this: 
tools: mcpClient.tools(),
Also make sure url is the tenant mcpHttpUrl, not the Corsair API base URL: 
// Good
url: conn.mcpHttpUrl

// Wrong
url: "https://api.corsair.dev"

​
Key usage

ValueUse
CORSAIR_API_KEYDeveloper API key for createClient()
CORSAIR_MCP_API_KEYTenant MCP secret for createVercelAiMcpClient()
The tenant MCP secret comes from: 
const key = await tenant.mcpKeys.create("vercel-ai-agent");

console.log(key.mcpHttpUrl);
console.log(key.secret);