danny-tdoan/realestate-rag-mcp

RealEstate MCP

What problem is it solving?

Real estate sites in Australia typically offer rigid, suburb-based filters that don’t match how people actually search. For example, a query like “renovated 3-bedroom homes in the Box Hill High School zone” simply isn’t possible.

This project leverages LLMs and MCP to enable flexible, natural language property search.

It allows users to search and ask questions about Australian real estate using publicly available property data. The project provides tools and runs a FastAPI server to serve a natural language property search API.

The system includes:

1. An indexing pipeline that downloads property data and populates a vector database for semantic search
2. An MCP server with different search tools
3. A demo client that sends queries and display results

Examples

Enter your query (or type 'quit' to exit): list 3 renovated houses with 4 bedrooms in Camberwell high school zone

Here are some renovated houses with 4 bedrooms located in the Camberwell High School zone:

1. **3/2-4 Georgina Parade, Camberwell VIC 3124**
   - Immaculately renovated throughout, this property boasts sleek modern style in a peaceful and private location. It's close to Hartwell Shopping Centre, Willison Park, and public transport.

2. **2/22 Russell Street, Camberwell VIC 3124**
   - This single-level boutique residence has been stylishly refurbished. Located on a prestigious tree-lined street, it offers sun-drenched dimensions for luxurious living and is near Camberwell Junction.

3. **6 Donna Buang Street, Camberwell VIC 3124**
   - Positioned in a picturesque, tree-lined street, this property presents an opportunity to build a family home or develop further. It's conveniently located for access to local schools.

Each property offers unique features and is ideally situated within the Camberwell High School zone.

1. Download Data with run_ingest_pipeline

This script downloads property details and auction results for a given city and date range.

Usage:

python -m cli.run_ingest_pipeline <city> <start_date> <end_date>

• <city>: Name of the city (e.g., melbourne)
• <start_date>: Start date in YYYY-MM-DD format (e.g., 2025-02-01)
• <end_date>: End date in YYYY-MM-DD format (e.g., 2025-05-10)

Example:

python -m cli.run_ingest_pipeline melbourne 2025-02-01 2025-05-10

This will download property data for Melbourne between February 1, 2025 and May 10, 2025.

2. Populate the Vector Database with populate_collection

This script populates a Chroma vector database collection with property details for semantic search.

Usage:

python -m cli.populate_collection.py <collection_name> <property_details_path>

• <collection_name>: Name of the ChromaDB collection to populate (e.g., melbourne)
• <property_details_path>: Path to the property details JSONL file (e.g., downloaded_data/property_details.jsonl)

Example:

python -m cli.populate_collection.py melbourne downloaded_data/property_details.jsonl

3. Run the server and the demo client

python -m cli.server

and

python -m cli.client