🍔 Zomato Voice Bot (MCP-Powered)

An interactive, voice-activated food ordering bot powered by the Model Context Protocol (MCP 2.x). Users can speak naturally to search restaurants, build food carts, handle out-of-stock item replacements interactively, calculate the best discounts/coupons, and confirm the exact final best amount before placing the order.

🌟 Key Features

1. Model Context Protocol (MCP 2.x) Integration:

   • Implements standard MCP tools:

     • search_restaurants: Discover restaurants by cuisine, rating, budget, and location.

     • get_menu: Browse categories, dishes, prices, and food details.

     • check_item_availability: Real-time stock verification with automatic alternative recommendations.

     • add_to_cart: Add items with quantity and customization options.

     • calculate_best_amount: Evaluates all coupons (ZOMATO50, FEAST150, GOLDDELIVERY, JUMBO200), maximizes discounts, calculates delivery fees, taxes, and generates the voice confirmation phrase.

     • confirm_and_place_order: Places order only after final user agreement, generating Order ID and payment QR.

     • track_order: Live order tracking and delivery partner status.

   • Dual-mode support: Runs with the built-in local MCP server or connects to the remote Zomato MCP endpoint (https://mcp-server.zomato.com/mcp).

2. Interactive Voice Assistance (STT & TTS):

   • Speech Recognition: Voice input via Web Speech API in the browser or terminal input.

   • Natural Voice Feedback: Spoken audio replies (English with Indian accent en-IN / gTTS / SpeechSynthesis).

   • Visual Voice Orb: Pulsing glowing voice orb with animated sound rings and waveform bars reacting to speech.

3. Smart Missing & Out-of-Stock Handling:

   • If an item is unavailable (e.g. Garlic Naan at Paradise Biryani or Cheesy Dip at Domino's), the bot does not fail or drop it silently.

   • It actively speaks:

     "However, Garlic Naan is currently out of stock at Paradise Biryani. They have Butter Naan (₹45) and Tandoori Roti (₹30) available. Would you like me to add one of these instead, or skip it?"

   • Listens to the user's answer ("Yes, add butter naan instead" or "Skip it") and updates the order state seamlessly.

4. Mandatory Final Best Amount Confirmation:

   • The bot automatically evaluates all coupons, applies the biggest discount, and speaks the exact final amount:

     "You have 2x Chicken Dum Biryani and 1x Butter Naan from Paradise Biryani. The subtotal is ₹605. With coupon FEAST150 applied saving ₹150, including delivery and taxes, your final best amount is ₹497. Should I confirm and place your order?"

   • Waits for the user's voice confirmation ("Yes, place order").

5. Production-Grade Persistence & Agent Memory:

   • Database Architecture: Async SQLAlchemy 2.0 with high-throughput SQLite (journal_mode=WAL) or PostgreSQL.

   • Persistent Active Cart: Food items and pricing breakdown remain intact across browser tabs, reloads, and server restarts.

   • Long-term Agent Memory: Autonomously learns dietary preferences (Veg/Non-Veg), preferred payment methods (UPI/Zomato Money/COD), favorite restaurants, and favorite dishes across voice sessions.

   • Order History & Audit Trail: Real-time order records with tracking URLs and payment states (including live order #8585879147).

   • MCP Tool Call Auditing: Every tool invocation is logged with execution duration (duration_ms), arguments, and status in mcp_audit_logs.

6. Modern Web Voice Console & Terminal CLI:

   • Web UI (http://localhost:8000): Full-screen foodie dark theme with live Cart sidebar, Best Offer badge, real-time MCP tool call inspector, Orders History modal, and Memory Profile manager.

   • Terminal CLI (./run.sh cli): Lightweight command-line interface with spoken audio.

Related MCP server: Uber Eats MCP Server

🚀 Quick Start

1. Run the Web Voice Assistant

./run.sh
# or
./run.sh web

Open your browser at http://localhost:8000. Tap the microphone icon or say:

"Order 2 chickpea salads from Lulu Hypermarket"

2. Run the Terminal Voice Assistant

./run.sh cli

3. Run the Automated Test Suite

./run.sh test
# or
.venv/bin/pytest -v tests/

4. Run the Zomato MCP Server directly via stdio

./run.sh mcp

🔌 Production REST API Endpoints

• GET /health: Production health check reporting uptime, database connectivity, and MCP server status.

• GET /api/orders: Retrieves historical and current orders with items, pricing, and live tracking links.

• GET /api/memory: Fetches long-term user preferences, favorite dishes, and dietary habits.

• POST /api/memory/preference: Explicitly updates dietary, payment, or restaurant preferences.

• GET /api/audit/logs: Audits all MCP tool executions with millisecond latency timings.

• GET /api/conversation/{session_id}: Retrieves chat history for multi-turn session persistence.

🗣️ Example Voice Walkthrough

⚙️ Configuration (.env)

# Database URL (SQLite WAL default or PostgreSQL)
DATABASE_URL=sqlite+aiosqlite:///./zomato_bot.db

# Optional: Gemini API Key
GEMINI_API_KEY=
GEMINI_MODEL=gemini-2.5-flash

# MCP Mode: "local" or "remote"
ZOMATO_MCP_MODE=remote
ZOMATO_MCP_REMOTE_URL=https://mcp-server.zomato.com/mcp
ZOMATO_MCP_AUTH_TOKEN=your_token_here

📂 Project Structure

zomato_bot/
├── zomato_bot/
│   ├── config.py             # App, DB & MCP configuration
│   ├── audio.py              # CLI text-to-speech audio engine
│   ├── cli.py                # Terminal voice ordering console
│   ├── db/
│   │   ├── session.py        # Async engine, connection pool & WAL pragmas
│   │   └── models.py         # SQLAlchemy 2.0 models (Session, Cart, Order, Memory, Audit)
│   ├── memory/
│   │   └── service.py        # Long-term preference learning & persistent cart/order CRUD
│   ├── mcp/
│   │   ├── server.py         # Official MCP 2.x Server implementation
│   │   ├── client.py         # MCP Client with async tool auditing
│   │   ├── tools.py          # Restaurant, menu, cart & checkout logic
│   │   └── data.py           # Restaurant, menu, coupon data
│   ├── agent/
│   │   ├── bot.py            # Unified agent interface
│   │   ├── nlp_fallback.py   # Multi-turn voice conversation & intent manager
│   │   └── prompts.py        # System prompt & voice dialogue rules
│   └── web/
│       ├── app.py            # FastAPI REST & WebSocket endpoints with lifespan
│       ├── templates/
│       │   └── index.html    # Interactive Voice UI with Orders & Memory modals
│       └── static/
│           ├── css/style.css # Zomato dark theme styling
│           └── js/
│               ├── voice.js  # Web Speech API STT/TTS engine
│               └── app.js    # UI, persistence syncing, Orders & Memory handlers
├── tests/                    # 16-test suite covering MCP, Agent, API & DB Persistence
├── run.sh                    # Launcher script
└── README.md