mcp

This MCP server exposes the Vapi MCP integration, enabling you to connect to Vapi's hosted or local MCP endpoints and manage assistants, calls, phone numbers, and tools through the MCP protocol. You can run the server locally and point your MCP client to the local node or to a remote instance, depending on your deployment. The server supports standard MCP operations such as listing, creating, updating, and deleting assistants, initiating outbound calls, and managing call histories. It also provides a set of utility commands for authentication (OAuth flow) and for discovering or invoking specific tools integrated with your Vapi workspace. Use the included tooling to orchestrate voice assistants and call workflows within your preferred MCP client or your own orchestrator.

Prerequisites:

• Node.js and npm installed on your system
• Access to a terminal/command prompt

1. Install dependencies and build the server

npm install
npm run build

2. Run the MCP server locally (example)

# Ensure your VAPI_TOKEN is available if required by your setup
export VAPI_TOKEN=<your_vapi_token>
node <path>/dist/index.js

3. (Optional) Configure via an MCP client

• Use the Local Development Config as a reference to point an MCP client at the local server:

{
  "mcpServers": {
    "mcp": {
      "command": "node",
      "args": ["<path>/dist/index.js"],
      "env": {
        "VAPI_TOKEN": "<your_vapi_token>"
      }
    }
  }
}

4. Build for production (optional)

npm run build
# Deploy the built dist/index.js to your production environment

Tips and notes:

• The server expects a VAPI_TOKEN if you’re connecting to Vapi services requiring authentication. Replace the placeholder with your actual token or configure environment accordingly.
• If you’re running in a cloud/container environment, ensure port mappings and environment variables are correctly passed to the runtime.
• The MCP interface supports common actions like vapi_list_assistants, vapi_create_call, vapi_update_assistant, etc. Use your MCP client to call these actions as documented in the Available Tools section of the README.
• For remote usage, you can point clients to the hosted MCP server at https://mcp.vapi.ai/mcp with the appropriate Authorization header.
• When testing, you can leverage the local dev config to simulate production behavior before deployment.