MCP 服务器

Vapi MCP Server

Enables integration with Vapi's voice AI platform to manage assistants, create and schedule phone calls with dynamic variables, and manage phone numbers through function calling.

README

Vapi MCP Server

The Vapi Model Context Protocol server allows you to integrate with Vapi APIs through function calling.

Claude Desktop Setup

Open Claude Desktop and press CMD + , to go to Settings.
Click on the Developer tab.
Click on the Edit Config button.
This will open the claude_desktop_config.json file in your file explorer.
Get your Vapi API key from the Vapi dashboard (https://dashboard.vapi.ai/org/api-keys).
Add the following to your claude_desktop_config.json file. See here for more details.
Restart the Claude Desktop after editing the config file.

Local Configuration

{
  "mcpServers": {
    "vapi-mcp-server": {
      "command": "npx",
      "args": [
          "-y",
          "@vapi-ai/mcp-server"
      ],
      "env": {
        "VAPI_TOKEN": "<your_vapi_token>"
      }
    }
  }
}

Remote Configuration

{
  "mcpServers": {
    "vapi-mcp": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.vapi.ai/mcp",
        "--header",
        "Authorization: Bearer ${VAPI_TOKEN}"
      ],
      "env": {
        "VAPI_TOKEN": "<your_vapi_token>"
      }
    }
  }
}

Example Usage with Claude Desktop

Create or import a phone number using the Vapi dashboard (https://dashboard.vapi.ai/phone-numbers).
Create a new assistant using the existing 'Appointment Scheduler' template in the Vapi dashboard (https://dashboard.vapi.ai/assistants).
Make sure to configure Claude Desktop to use the Vapi MCP server and restart the Claude Desktop app.
Ask Claude to initiate or schedule a call. See examples below:

Example 1: Request an immediate call

I'd like to speak with my ShopHelper assistant to talk about my recent order. Can you have it call me at +1234567890?

Example 2: Schedule a future call

I need to schedule a call with Mary assistant for next Tuesday at 3:00 PM. My phone number is +1555123456.

Example 3: Make a call with dynamic variables

I want to call +1234567890 with my appointment reminder assistant. Use these details:
- Customer name: Sarah Johnson
- Appointment date: March 25th
- Appointment time: 2:30 PM
- Doctor name: Dr. Smith

Using Variable Values in Assistant Prompts

The create_call action supports passing dynamic variables through assistantOverrides.variableValues. These variables can be used in your assistant's prompts using double curly braces: {{variableName}}.

Example Assistant Prompt with Variables

Hello {{customerName}}, this is a reminder about your appointment on {{appointmentDate}} at {{appointmentTime}} with {{doctorName}}.

Default Variables

The following variables are automatically available (no need to pass in variableValues):

{{now}} - Current date and time (UTC)
{{date}} - Current date (UTC)
{{time}} - Current time (UTC)
{{month}} - Current month (UTC)
{{day}} - Current day of month (UTC)
{{year}} - Current year (UTC)
{{customer.number}} - Customer's phone number

For more details on default variables and advanced date/time formatting, see the official Vapi documentation.

Remote MCP

To connect to Vapi's MCP server remotely:

Streamable HTTP (Recommended)

The default and recommended way to connect is via Streamable HTTP Transport:

Connect to https://mcp.vapi.ai/mcp from any MCP client using Streamable HTTP Transport
Include your Vapi API key as a bearer token in the request headers
Example header: Authorization: Bearer your_vapi_api_key_here

SSE (Deprecated)

Server-Sent Events (SSE) Transport is still supported but deprecated:

Connect to https://mcp.vapi.ai/sse from any MCP client using SSE Transport
Include your Vapi API key as a bearer token in the request headers
Example header: Authorization: Bearer your_vapi_api_key_here

This connection allows you to access Vapi's functionality remotely without running a local server.

Development

# Install dependencies
npm install

# Build the server
npm run build

# Use inspector to test the server
npm run inspector

Update your claude_desktop_config.json to use the local server.

{
  "mcpServers": {
    "vapi-local": {
      "command": "node",
      "args": [
        "<path_to_vapi_mcp_server>/dist/index.js"
      ],
      "env": {
        "VAPI_TOKEN": "<your_vapi_token>"
      }
    },
  }
}

Testing

The project has two types of tests:

Unit Tests

Unit tests use mocks to test the MCP server without making actual API calls to Vapi.

# Run unit tests
npm run test:unit

End-to-End Tests

E2E tests run the full MCP server with actual API calls to Vapi.

# Set your Vapi API token
export VAPI_TOKEN=your_token_here

# Run E2E tests
npm run test:e2e

Note: E2E tests require a valid Vapi API token to be set in the environment.

Running All Tests

To run all tests at once:

npm test

References

Supported Actions

The Vapi MCP Server provides the following tools for integration:

Assistant Tools

list_assistants: Lists all Vapi assistants
create_assistant: Creates a new Vapi assistant
update_assistant: Updates an existing Vapi assistant
get_assistant: Gets a Vapi assistant by ID

Call Tools

list_calls: Lists all Vapi calls
create_call: Creates an outbound call with support for:
- Immediate or scheduled calls
- Dynamic variable values through assistantOverrides
get_call: Gets details of a specific call

Note: The create_call action supports scheduling calls for immediate execution or for a future time. You can also pass dynamic variables using assistantOverrides.variableValues to personalize assistant messages.