Vapi’s MCP Server is Live on Vercel

We’ve partnered with Vercel to launch our MCP server, now supporting streamable-HTTP for more reliable connections. 

What’s New

• Expose Vapi as Tools: Any LLM, agent framework, or AI app that supports MCP (e.g., Claude, Cursor, LangChain) can now access all of Vapi’s endpoints as tools.
• Built for Real-Time: Built on Vercel with streamable HTTP for more reliable connections - no more issues with long-lived SSE connections.

What is an MCP Server?

An MCP server is a lightweight program that exposes tools and real-world actions to AI models via the Model Context Protocol.

Think of it as a universal toolbox - LLMs and agent frameworks can call your server to:

• Make phone calls
• Fetch transcripts and call logs
• Build workflows
• Analyze call data

Use Cases

With Vapi’s MCP server on Vercel, your AI agents can:

• Initiate inbound or outbound calls
• Programmatically write and run test cases for your assistants
• Pull call logs/transcripts and instantly analyze with Claude
• Write multi-step workflows to navigate your assistant through logical steps using our workflows API 
• Integrate Vapi with your own servers or custom frameworks

Why Streamable-HTTP?

Server-Sent Events (SSE) require long-lived connections that can often break at scale, especially when handling concurrent clients.

Streamable HTTP is stateless and more reliable, with each request handling its own lifecycle.

Get Started

• MCP docs
• MCP server
• Dashboard
• Check out Vercel’s blog post on details for how we hosted it