voice-agent

Voice Agent MCP Server integrates Twilio Voice, Deepgram, and OpenAI to enable AI-driven voice workflows for HR tasks such as conducting candidate interviews, delivering results, and outreach. The server exposes MCP tools that Claude or other MCP clients can call to initiate calls, manage interviews, and send notifications. Real-time audio is streamed via Twilio Media Streams and processed through Deepgram with an OpenAI-backed prompt system to generate natural, context-aware conversations. Tools include conducting interviews, notifying candidates of results, discussing job openings, checking call status, and verifying Twilio configuration. To use it, configure the MCP server in Claude Desktop (or your MCP client) with the provided mcp-config.json entry, ensure environment variables are set, and start the web server. Then invoke the available tools in your conversations to trigger voice actions such as starting an interview or sending a notification.

Prerequisites:

• Node.js 22+ installed
• npm installed
• A Twilio account with Account SID, Auth Token, and a configured outbound number
• A Deepgram account with an API key
• A public URL (ngrok or production URL) for webhooks

Installation steps:

1. Clone the repository and install dependencies

git clone https://github.com/prakharbhardwaj/voice-agent-mcp-server.git
cd voice-agent-mcp-server
npm install

2. Create and configure environment variables

# Create a .env or set variables in your environment
PORT=3000
SERVER_URL=https://your.public-url
TWILIO_ACCOUNT_SID=your_twilio_account_sid
TWILIO_AUTH_TOKEN=your_twilio_auth_token
TWILIO_PHONE_NUMBER=+your_twilio_phone_number
DEEPGRAM_API_KEY=your_deepgram_api_key

3. Configure MCP server entry

• Edit mcp-config.json (or the equivalent MCP manifest) to include your server:

{
  "mcpServers": {
    "voice-agent-mcp-server": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/your/voice-agent-mcp-server/mcp-server.js"],
      "env": {
        "NODE_ENV": "production",
        "SERVER_URL": "your_ngrok_url_or_server_url",
        "TWILIO_ACCOUNT_SID": "your_twilio_account_sid",
        "TWILIO_AUTH_TOKEN": "your_twilio_auth_token",
        "TWILIO_PHONE_NUMBER": "+your_twilio_phone_number"
      }
    }
  }
}

4. Start the server

npm run dev

5. Ensure your MCP client (e.g., Claude Desktop) is configured to connect to the MCP server using the provided configuration and that the public URL in SERVER_URL is reachable.

Tips and common issues:

• Ensure SERVER_URL is publicly accessible and reachable from Twilio and Deepgram callbacks.
• Keep all credentials in environment variables; avoid committing .env files.
• If Claude desktop does not load the MCP server, verify absolute paths in mcp-config.json and restart Claude Desktop after changes.
• Review Deepgram and Twilio logs for connection or authorization errors.
• Use the supplied tools list to extend functionality by adding new tools in src/mcp/tools.js and wiring prompts in src/mcp/prompts.js.