Facebook Marketplace Scraper

A local web tool to scrape Facebook Marketplace search/category pages, auto-scroll to collect many results (even when logged in), and extract structured listing data using either an LLM or a fast programmatic parser.

Features

• Optional login: type-like-human into name="email" and name="pass", then submit just like a user
• Headful debug mode (watch the browser) or headless mode for background runs
• Smart auto-scroll with cumulative item tracking (works with virtualized feeds)
• Desired item target: stop as soon as we’ve collected N unique listings
• Notification prompt suppression/dismissal after login to prevent blocking
• Two extraction modes:
  • LLM (Gemini): high-quality structured JSON
  • Parser (Cheerio): fast, no-LLM, DOM-based extraction
• Simple local web UI with fields for query/category, location, radius, price range, days since listed, sort exact/newest, optional login, desired items, and debug

Requirements

• Node.js 18+
• For LLM mode: Google Gemini API key (GOOGLE_API_KEY or GEMINI_API_KEY)

Install

npm install

Environment

Create a .env (optional) or export env vars:

• GOOGLE_API_KEY or GEMINI_API_KEY (required for LLM mode)
• GEMINI_MODEL (optional): defaults to gemini-2.5-pro
• PORT (optional): defaults to 3000
• SCRAPE_BASE_URL (optional): defaults to https://www.facebook.com
• PUPPETEER_SKIP_DOWNLOAD (optional): set true to skip Chromium auto-download

Run the Web UI (local only)

npm start

Open http://localhost:3000, fill out the form, and click Run.

Form fields include:

• Query or Category (one required)
• Location (city slug or numeric ID), radius
• Min/Max Price, Days Since Listed, Sort newest, Exact match
• Extraction Mode: AI (LLM) or Parser
• Email and Password (optional) to log in first
• Desired Items (stop after N unique items)
• Show browser (debug) to watch and log the run

Example of the URL it builds:

https://www.facebook.com/marketplace/calgary/search?query=<encoded>&daysSinceListed=<int>&exact=false

API (optional)

POST /api/scrape

Body (JSON):

{
  "query": "iphone",
  "category": "bicycles",
  "location": "calgary",
  "locationId": 102190403,
  "radius": 200,
  "minPrice": 100,
  "maxPrice": 500,
  "daysSinceListed": 7,
  "exact": true,
  "sortBy": "creation_time_descend",
  "mode": "programmatic",
  "desiredItemCount": 60,
  "email": "you@example.com",
  "password": "yourPassword123",
  "debug": true
}

Response:

{ "htmlLength": 123456, "json": "{\"listings\":[...]}" }

How extraction works

Programmatic Parser

• Parses a container (or synthetic container) after scrolling completes
• Walks each listing a[href*="/marketplace/item/"] and collects:
  • url (made absolute using base URL)
  • image_url (first img inside)
  • name (from img[alt], with trailing location stripped)
  • price ($.., CA$.., or Free from anchor text)
• Dedupes by URL (or name+image fallback)

LLM (Gemini)

• Builds a compact JSON context of candidate items and asks the model to emit { listings: [...] } with the same fields as above

Troubleshooting

• Login fills the email/password fields and submits using Enter on password, then button click fallbacks. If your login UI differs, update the selectors.
• If a notification dialog appears post-login and blocks scrolling, it’s auto-suppressed; if you see a new variation, add the button text to the dismiss list.
• Use the “Show browser (debug)” checkbox to watch the run and view detailed logs, including scrolling and typing.
• If Chromium download fails during install, set PUPPETEER_SKIP_DOWNLOAD=true and ensure a compatible Chrome/Chromium is available on your system.

Legal/Usage Notice

• This tool is for educational and personal use only.
• Ensure you have permission and comply with all applicable laws, site Terms of Service, and robots.txt directives.
• The authors are not responsible for any unauthorized use or damages. Use at your own risk.