Gemini Live Voice Agent + MCP

A demo project showcasing a realtime salon voice assistant with a React frontend, a Python FastAPI relay, and a Node.js MCP server.
It demonstrates Gemini Live API voice interaction with MCP tool calling for salon workflows like product lookup and appointment management.

Tech stack

• Frontend: React + TypeScript + Vite
• Backend: Python + FastAPI + google-genai
• MCP Server: Node.js + TypeScript + Express + MCP SDK
• Database: SQLite
• AI: Gemini Live API

Required versions

• Python: 3.12+
• Node.js: 20+

Clone and setup

git clone <repo-url>
cd gemini_voice_agent

MCP server setup

cd mcp-server
npm install

Create mcp-server/.env from example:

cp .env.example .env

Seed the MCP SQLite database (uses mcp-server/seed.ts):

npm run seed

This resets and repopulates demo products and appointments data.

Run MCP server:

npm run dev

MCP endpoint:

http://localhost:8090/mcp

Python backend setup

cd backend

Create backend/.env from example and set your Gemini key:

cp .env.example .env

Required key in backend/.env:

GEMINI_API_KEY=your_gemini_api_key_here

Install dependencies and run backend:

pip install -e .
python main.py

Backend endpoints:

http://localhost:8000/health
ws://localhost:8000/ws

Frontend setup

cd frontend
npm install

Create frontend/.env from example:

cp .env.example .env

Run frontend:

npm run dev

Default frontend URL:

http://localhost:5173

Quick run order

1. Start mcp-server (npm run dev)
2. Start backend (python main.py)
3. Start frontend (npm run dev)

Env examples kept in repo

• backend/.env.example
• frontend/.env.example
• mcp-server/.env.example