If you’ve ever missed a great local deal because you weren’t online at the right time, you know the pain. I set out to solve that with a small, reliable, and privacy-respecting script that watches Facebook Marketplace for you, filters the noise, and pings your phone only when the right items appear.

The Problem

• Marketplace deals move fast; manual refreshing is tedious.
• Search results include lots of noise: wrong items, overpriced listings, accessories instead of the product.
• You want immediate, actionable alerts on your phone with a link to buy.

The Solution

I built a Python service that:

• Periodically queries Facebook Marketplace for specific searches.
• Parses listings from the page’s embedded JSON (with an HTML fallback).
• Stores everything in SQLite to avoid repeat work/alerts.
• Filters items using per-query rules (like max price).
• Sends a push notification to your iPhone via Pushover or to any device via ntfy.
• Optionally runs AI image checks (Gemini) before alerting to ensure the listing actually shows the product you want.

Why This Is Handy

• Saves time: no more constant refreshing.
• Reduces noise: alerts are gated by price and optional image verification.
• Faster response: being first boosts your odds of snagging good deals.
• Portable: run it locally or deploy to a VPS and forget it.

Key Features

• Config-driven searches in config.yaml with per-query filters and AI prompts.
• APScheduler-based periodic jobs or a one-shot CLI mode (--once).
• SQLite persistence (dedupe + skip logic for already-seen items).
• Notifications:
• Pushover: iOS-friendly, battle-tested.
• ntfy: simple, self-hostable, works across devices.
• AI image filter (Gemini): only alert if the main image looks like the actual product (e.g., Xbox One S/X).
• Robust scraping:
• Primary parsing from Marketplace’s embedded JSON.
• Fallback to HTML anchors for resilience.
• Debuggable:
• LOG_LEVEL=DEBUG to see fetch timing, parse counts, filter decisions.
• Proxy support (e.g., http://localhost:8080) for traffic inspection.

Technical Highlights

• Per-query max_price filtering and ai_prompt overrides.
• JSON-first parsing from the app’s application/json script blocks.
• AI prompt is configurable at the global or per-query level.
• Rate-limit backoff for Gemini: reads “retry in Xs” and waits before retrying.
• Only new items trigger notifications, and only if AI passes (when enabled).
• Every listing—pass or fail—is stored once with ai_pass and ai_model so it’s never reprocessed.

Privacy & Ethics

• You supply your own session cookies (optionally) and API keys.
• No external storage of secrets; everything is on your machine or server.
• Respect the site’s terms and your local laws. Use responsibly.

Deploy It Anywhere

• Local dev: python main.py --once to test, then python main.py for scheduled runs.
• Server: systemd service for reliable background execution on Oracle Linux (or nohup, tmux).

Real-World Benefits

• Find underpriced items fast (e.g., “Xbox One under $50”).
• Avoid accessory-only listings through AI image checks.
• Keep a clean history of seen items and outcomes for auditing.

Future Ideas

• Multi-image analysis and “any image qualifies” logic.
• Region-aware price normalization.
• Web UI for logs, status, and manual triage.
• Slack/Discord notification channels.

How To Use (Quick Start)

• Set up Python venv and install requirements.
• Add your searches, headers, cookies, and notification settings to config.yaml.
• Optional: add Gemini gemini_api_key and customize the prompt.
• Run once to test:
• LOG_LEVEL=DEBUG python main.py --once
• Run scheduled
• python main.py (or set up a systemd service on your server).

Configuration

Edit `config.yaml` (example below). You can also use JSON, but YAML is recommended.

queries:
  - query: "xbox one"
    max_price: 50
    ai_prompt: "Does this photo clearly show a Microsoft Xbox One S or Xbox One X console? Respond with only 'YES' or 'NO'."

check_interval_minutes: 10

database:
  path: "marketplace.db"

notification:
  service: "ntfy"  # "pushover" or "ntfy"
  pushover:
    api_token: "YOUR_PUSHOVER_APP_TOKEN"
    user_key: "YOUR_PUSHOVER_USER_KEY"
  ntfy:
    server: "https://ntfy.sh"  # or your self-hosted server
    topic: "your-topic-name"

facebook:
  headers:
    user_agent: "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0.0.0 Safari/537.36"
  # You can paste a full raw Cookie header string or a key:value map
  cookies: "c_user=...; xs=...; fr=...; datr=...; sb=..."
  proxy: "http://localhost:8080"  # optional; useful with tools like Burp

# Enable verbose item dump for debugging
debug:
  dump_parsed_items: false

# AI image filter (optional) - filters notifications to qualifying images
ai:
  gemini_api_key: "YOUR_GEMINI_API_KEY"  # or set env var GEMINI_API_KEY
  model: "gemini-2.5-flash"
  prompt: "Does this photo clearly show a Microsoft Xbox One S or Xbox One X console? Respond with only 'YES' or 'NO'."
```

Closing Thoughts

This project turns Marketplace hunting from a manual chore into an automated assistant that works for you. It’s small, configurable, and designed to be extended. Whether you’re chasing a specific console, bike size, or niche tool, you get relevant, timely alerts—and more wins.

If you are interested in the code, send me a message.