Give a Vercel AI SDK app 100+ tools with MCP

npm install ai @ai-sdk/mcp @ai-sdk/openai

Create a free APIStack account and set your key as APISTACK_API_KEY.

import { createMCPClient } from '@ai-sdk/mcp'
import { generateText, stepCountIs } from 'ai'
import { openai } from '@ai-sdk/openai'

const apistack = await createMCPClient({
  transport: {
    type: 'http',
    url: 'https://apistack.online/api/mcp/connect',
    headers: { 'x-api-key': process.env.APISTACK_API_KEY }
  }
})

try {
  const { text, steps } = await generateText({
    model: openai('gpt-5-mini'),
    tools: await apistack.tools(),
    stopWhen: stepCountIs(6),
    prompt: 'Is jane.doe@gmial.com a valid email address to send a receipt to? Check it with APIStack.'
  })
  console.log(steps.flatMap(s => s.toolCalls.map(c => c.toolName)))
  console.log(text)
} finally {
  await apistack.close()
}

Output from a real run (ai 7.0, @ai-sdk/mcp 2.0, gpt-5-mini):

[ 'search_apis', 'get_api_details', 'call_api' ]
Short answer: No — the validator reports jane.doe@gmial.com is not valid for sending a receipt.
- MX (mail exchanger) records: none found → likely undeliverable
- Typo suggestion: gmail.com (high confidence)
- Risk score: 60 (level: high)
...

If your app only needs a few APIs, add ?apis=email-address-validator,phone-number-validation (or ?categories=finance) to the URL. You then get one tool per API with its own schema, which is faster for the model than searching.

The same tools work with streamText in a Next.js route handler. Create the client per request and close it in onFinish, or reuse one client across requests in a long-running server.