import argparse import datetime from homeharvest import scrape_property def main(): parser = argparse.ArgumentParser(description="Home Harvest Property Scraper") parser.add_argument( "location", type=str, help="Location to scrape (e.g., San Francisco, CA)" ) parser.add_argument( "-l", "--listing_type", type=str, default="for_sale", choices=["for_sale", "for_rent", "sold"], help="Listing type to scrape", ) parser.add_argument( "-o", "--output", type=str, default="excel", choices=["excel", "csv"], help="Output format", ) parser.add_argument( "-f", "--filename", type=str, default=None, help="Name of the output file (without extension)", ) parser.add_argument( "-p", "--proxy", type=str, default=None, help="Proxy to use for scraping" ) parser.add_argument( "-d", "--days", type=int, default=None, help="Sold/listed in last _ days filter.", ) parser.add_argument( "-r", "--radius", type=float, default=None, help="Get comparable properties within _ (eg. 0.0) miles. Only applicable for individual addresses.", ) parser.add_argument( "-m", "--mls_only", action="store_true", help="If set, fetches only MLS listings.", ) args = parser.parse_args() result = scrape_property( args.location, args.listing_type, radius=args.radius, proxy=args.proxy, mls_only=args.mls_only, last_x_days=args.days, ) if not args.filename: timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S") args.filename = f"HomeHarvest_{timestamp}" if args.output == "excel": output_filename = f"{args.filename}.xlsx" result.to_excel(output_filename, index=False) print(f"Excel file saved as {output_filename}") elif args.output == "csv": output_filename = f"{args.filename}.csv" result.to_csv(output_filename, index=False) print(f"CSV file saved as {output_filename}") if __name__ == "__main__": main()