MCP-powered voice agent

This project implements a voice agent that combines web search capabilities via Firecrawl with Supabase database operations through MCP (Model Context Protocol).

Installation

Ensure you have Python 3.x installed and run:

pip install -r requirements.txt

Implementation: agent.py

This implementation uses AssemblyAI's services for speech-to-text, along with Firecrawl for web search and Supabase for database operations.

Requirements

• Firecrawl API key
• Supabase access token
• OpenAI API key
• AssemblyAI API key
• LiveKit credentials

Setup

Copy .env.example to .env and configure the following environment variables:

FIRECRAWL_API_KEY=your_firecrawl_api_key
SUPABASE_ACCESS_TOKEN=your_supabase_token
OPENAI_API_KEY=your_openai_api_key
ASSEMBLYAI_API_KEY=your_assemblyai_api_key
LIVEKIT_URL=your_livekit_url
LIVEKIT_API_KEY=your_livekit_api_key
LIVEKIT_API_SECRET=your_livekit_api_secret

Running

Start the agent using:

python agent.py

The agent will:

1. Connect to LiveKit
2. Initialize the MCP server for Supabase integration
3. Set up voice interaction capabilities
4. Start listening for user input

Features

• Real-time web search using Firecrawl
• Supabase database integration via MCP
• Voice interaction capabilities:
  • Silero VAD (Voice Activity Detection)
  • AssemblyAI Speech-to-Text
  • OpenAI GPT-4 for language processing
  • OpenAI TTS for text-to-speech

Contribution

Contributions are welcome! Feel free to fork this repository and submit pull requests with your improvements.