MCP Integrations
Connect your application or agent to a Model Context Protocol (MCP) server to use its tools. Choose an integration based on how you build your agent:
| Integration | Use it when | Connection setup |
|---|---|---|
| eve | You build durable backend agents with eve. | Define a connection under agent/connections/ and configure authentication. |
| AI SDK | You build agents and model calls with the AI SDK. | Create an MCP client and pass its tools to the model. |
| TanStack AI | You use TanStack AI's chat and streaming APIs. | Pass an MCP client to chat() for managed tool discovery and cleanup. |
The AI SDK and TanStack AI examples connect to a server hosted on Vercel and send model requests through AI Gateway. The eve guide connects to an OAuth-protected MCP server through Vercel Connect.
Each integration needs the MCP server's endpoint and any credentials that server requires. Model credentials and MCP credentials authenticate separate requests.
For OAuth-protected servers, see the Vercel Connect guides for AI SDK and eve.
- Deploy an MCP server on Vercel to expose your application's tools.
- Use Vercel's MCP server to work with Vercel projects, deployments, and documentation.
Last updated September 15, 2026
Was this helpful?