HotelHive

HotelHive is a conversational hotel search assistant powered by LangGraph, LangChain, and an MCP tool server. It grounds responses in your local Excel datasets and supports follow-up questions with conversational memory.

Features

• Conversational agent with memory (multi-turn chat)
• MCP tool hotel_list backed by your Excel data
• Schema-free prompting: the assistant infers fields directly from the records
• Flexible queries by city/county/price/amenities/etc.
• Lists all matching hotels with optional pagination

Project Structure

• client.py: Runs the chat client, connects to the MCP server via stdio, loads tools, and maintains conversation history.
• server/hotelinfo_server.py: FastMCP server exposing the hotel_list tool, reading data from data/hotels.xlsx.
• agent/hotel_finder.py: Prompt + LLM runnable used by the server tool to generate grounded answers.
• data/: Excel files (hotels.xlsx, etc.) used as the knowledge source.
• type/: Pydantic models (not required in current schema-free flow).
• requirements.txt: Python dependencies.

Prerequisites

• Python 3.12+
• A Google Generative AI API key (API_KEY) and model name (MODEL, e.g., gemini-2.0-flash).

Setup

1. Create and activate a virtual environment:
   • python3 -m venv .venv
   • source .venv/bin/activate
2. Install dependencies:
   • pip install -r requirements.txt
3. Create a .env file in the project root with:
   • API_KEY=your_google_genai_key
   • MODEL=gemini-2.0-flash
4. Ensure your data file exists:
   • data/hotels.xlsx

Run

• From the project root, activate the venv and run:
  • python3 client.py
• Interact in the console (type exit to quit). Examples:
  • suggest hotels under 150 in chicago
  • list all hotels in miami
  • share details for hotel_85

Notes

• The assistant answers using only the provided Excel data; if a requested field doesn’t exist, it will say it’s not available.
• Large result sets will paginate (first 50 shown) and continue when you confirm.

Troubleshooting

• If you see import/module errors, ensure the client launches the server with python -m server.hotelinfo_server (already configured in client.py).
• If responses seem off, confirm your .env variables and that data/hotels.xlsx contains the expected columns and values.

License

• MIT