Introducing Vapi MCP Client

We are excited to launch Vapi MCP client. You can now build voice agents that immediately integrate with existing MCP servers to use their tools and context. 

Building voice agents often means connecting them to the outside world - scheduling appointments, updating CRMs, sending messages, or fetching real-time data. Traditionally, this required building and maintaining individual integrations for every service.
Vapi MCP changes that for good.

What's New

We're thrilled to announce that Vapi now natively supports the Model Context Protocol (MCP).
Vapi MCP is a protocol-level implementation that transforms how voice agents interact with external systems. Vapi handles JSON-RPC over HTTP/2, tool schema validation, and automatic retries, while your MCP server focuses on business logic

Why It Matters

With Vapi + MCP, you gain:

1. Easy Tool Integration: Connect your Vapi agent to MCP-compatible services like Zapier or Composio. Instantly access thousands of pre-built actions across countless apps without writing individual API integrations.
2. Context-Aware Capabilities: Tools aren't static. Vapi fetches the currently available tools from your configured MCP server when a call starts, allowing your agent's capabilities to adapt based on the conversation or server-side changes.
3. Reduced Development Complexity: Point Vapi to your MCP server URL, and let the protocol handle discovery and interaction. Focus your valuable engineering hours on core agent logic and user experience.

How It Works

The Vapi MCP integration follows a straightforward flow during a call:\

1. Connect & Discover: When a call starts, Vapi connects to the MCP server URL you configured. The server responds with a list of available tools and their capabilities.
2. Inject Context: Vapi adds these discovered tools to your assistant's available resources for the LLM.
3. Invoke & Execute: If the LLM decides to use an MCP-provided tool, Vapi sends the request to the MCP server.
4. Return Result: The MCP server executes the action and sends the outcome back to Vapi, informing the agent's next step.

Important Note: The MCP tool you configure in the Vapi Dashboard acts as a pointer to the server; the LLM interacts with the specific tools provided by that server, not the pointer itself.

Getting Started

1. Obtain An MCP Server URL:
   • Sign up or log in to an MCP provider (like Zapier MCP Servers or Composio) and generate/copy your unique MCP server URL.
   • Remember to treat this URL as a sensitive credential.
2. Create the MCP Tool in Vapi:
   • Log in to your Vapi Dashboard and go to the Tools page.
   • Click the Create Tool button and select MCP.
   • Provide a name (e.g., "Zapier") and paste your MCP server URL into the serverUrl field.
3. Add Tool to Your Assistant:

• Navigate to the Assistants page and select your assistant.
• Go to the Tools tab.
• Select your newly created MCP tool from the dropdown menu.
• Click Publish to save your changes.

Example Workflow

So let's say a customer support agent needs to check order statuses + reschedule appointments. After adding Zapier's MCP server URL to Vapi:

1. User says, "Check my recent order", then the agent auto-pulls Shopify data via Zapier
2. And if the user says, "Push delivery to Friday", the agent now updates both Shopify and Google Calendar in one voice command. Your assistant can now leverage the tools available through the configured MCP server.
3. Review our Vapi MCP Documentation for more detailed instructions and best practices.

Stay curious, keep iterating, and let's build something real together.