Vercel Connect

Vercel Connect manages credentials for apps and agents that call Tavily’s MCP server or REST API. Use OAuth when each user connects their own Tavily account, or an API key for shared application access. Connect handles credentials; Eve runs agents; AI Gateway routes model requests. You can use Connect without either. For Tavily tools used directly in the AI SDK, see Vercel AI SDK.

Prerequisites

A Vercel account and permission to create connectors and link projects.
A Tavily account with credits. API-key connections also need a Tavily API key.
Node.js 22 or newer for the scripts below, or Node.js 24 or newer for Eve, and the Vercel CLI.

Set up the connector

Choose MCP or API

Open Tavily on Vercel Connect, select Create Connector, and choose your team. Select MCP for agent tools or API for direct REST calls.

Vercel Connect offering MCP and API connections for Tavily

Configure authentication

For MCP OAuth, keep Managed, enter a name, and continue. Vercel manages the OAuth client. For the API connection, enter your key from the Tavily dashboard.

Managed OAuth configuration with connector name and optional UID

Create the connector

Leave the optional icon and colors as they are, or customize them. Select Create Connector, then Continue. Copy the connector’s UID for your code; the examples use tavily/your-connector-name as a placeholder.

Customization and completion screens

Optional connector branding and Create Connector button

Confirmation that the Tavily connector has been created

Link a project

In the connector’s Projects section, select Connect Project, choose your project, and enable Development for local testing. If it is already linked, check the enabled environments. A deployment is not required.

Tavily connector page with the Connect Project button — The Projects section of a Tavily connector; your connector name and UID will differ.

Prepare a local project

Use an existing project directory, or create one:

mkdir tavily-connect-example
cd tavily-connect-example
npm init -y
npm install @vercel/connect @ai-sdk/mcp
vercel login
vercel link
vercel env pull

Select the same project you linked to the connector. vercel env pull creates .env.local with a development VERCEL_OIDC_TOKEN. Keep this file out of Git and refresh it with the same command if the token expires. See Vercel’s authentication guide.

Example usage

Find three sources about a company’s product announcements, then use the results in a news feed, account brief, or research agent. Choose the example that matches your connector.

MCP with OAuth
REST with an API key
Eve agent

Save this as search.mjs. The first run prints an authorization link. Open it, sign in to Tavily, approve access, and rerun the script.

search.mjs

import {
  getToken,
  startAuthorization,
  UserAuthorizationRequiredError,
} from '@vercel/connect';
import { createMCPClient } from '@ai-sdk/mcp';

const connector = process.env.TAVILY_CONNECTOR_ID;
if (!connector) throw new Error('Set TAVILY_CONNECTOR_ID.');
const subject = { type: 'user', id: 'local-user' };

async function main() {
  let token;
  try {
    token = await getToken(connector, { subject });
  } catch (error) {
    if (!(error instanceof UserAuthorizationRequiredError)) throw error;
    const { url } = await startAuthorization(connector, { subject });
    console.log('Authorize Tavily, then rerun this script:', url);
    return;
  }

  const client = await createMCPClient({
    transport: {
      type: 'http',
      url: 'https://mcp.tavily.com/mcp',
      headers: { Authorization: `Bearer ${token}` },
    },
  });
  try {
    const result = await client.callTool({
      name: 'tavily_search',
      arguments: {
        query: 'Stripe agentic commerce product announcements',
        time_range: 'month',
        max_results: 3,
      },
    });
    if (result.isError) throw new Error('Tavily search failed.');
    console.log(JSON.stringify(result, null, 2));
  } finally {
    await client.close();
  }
}

await main();

TAVILY_CONNECTOR_ID='tavily/your-connector-name' node --env-file=.env.local search.mjs

The response contains search results with titles, URLs, and content snippets. Creating a connector does not authorize a user: the grant belongs to the same user ID used in both calls. In an application, replace local-user with the signed-in user’s ID from your server-side session.To let a model choose and call tools, follow Vercel’s AI SDK and MCP guide.

Use the UID of an API connector configured with a Tavily key. Save this as search-rest.mjs:

search-rest.mjs

import { getToken } from '@vercel/connect';

const connector = process.env.TAVILY_CONNECTOR_ID;
if (!connector) throw new Error('Set TAVILY_CONNECTOR_ID.');
const apiKey = await getToken(connector, { subject: { type: 'app' } });

const response = await fetch('https://api.tavily.com/search', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${apiKey}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    query: 'Stripe agentic commerce product announcements',
    time_range: 'month',
    max_results: 3,
  }),
});
if (!response.ok) throw new Error(`Tavily returned ${response.status}`);
const { results } = await response.json();
console.table(results.map(({ title, url }) => ({ title, url })));

TAVILY_CONNECTOR_ID='tavily/your-api-connector' node --env-file=.env.local search-rest.mjs

Connect supplies the stored key at runtime. This app-scoped call uses the shared credential and does not prompt each user to authorize Tavily.

Eve is Vercel’s agent framework. Create an agent with Node.js 24 or newer, then connect a model through Eve’s /login flow:

npx eve@latest init tavily-agent

Exit the terminal UI and run these commands inside tavily-agent. Select a Vercel project when prompted. For this example, create an MCP API-key connector and enter your Tavily key at the CLI prompt:

npm install @vercel/connect
vercel link
vercel connect create tavily --connection-method mcp-api-key --name tavily-agent
vercel connect attach 'tavily/your-connector-name' --environment development
vercel env pull

Replace the UID above and below with the UID returned by create. Add agent/connections/tavily.ts:

agent/connections/tavily.ts

import { defineMcpClientConnection } from 'eve/connections';
import { connect } from '@vercel/connect/eve';

export default defineMcpClientConnection({
  url: 'https://mcp.tavily.com/mcp',
  tools: { allow: ['tavily_search', 'tavily_extract'] },
  auth: connect({
    connector: 'tavily/your-connector-name',
    principalType: 'app',
  }),
});

Run npm run dev and ask:

Use Tavily to find three recent Stripe product announcements. Give each title, a one-sentence summary, the publication date if verified, and a source link.

This example shares one Tavily credential across the agent. For individual OAuth accounts, use an MCP OAuth connector and principalType: 'user' with an authenticated user principal. See Eve authorization with Connect.

Gateways and routers

Integrations

Partnerships

Machine Payments

Prerequisites

Set up the connector

Prepare a local project

Example usage

Resources

​Prerequisites

​Set up the connector

​Prepare a local project

​Example usage

​Resources

Prerequisites

Set up the connector

Prepare a local project

Example usage

Resources