Watch Arbitrage Tracker — Rolex/Patek/AP Cross-Platform Alerts

Apify Actor source code. A cross-platform price arbitrage tracker for professional watch dealers trading Rolex, Patek Philippe, and Audemars Piguet across Chrono24, The 1916 Company (formerly WatchBox), Bob's Watches, Watchfinder, European Watch Company, and Watches of Switzerland. Real-time Telegram alerts when a listing is mispriced versus the cross-platform median. Doubles as an MCP server for AI agents.

▶ Try it on Apify Store — runs in the cloud, free tier sufficient for most dealers.

---

What's in this repo

This repository is the public source for the Apify Actor kazkn/watch-arbitrage-mcp. The full product documentation, screenshots, input schema, dealer-facing tutorial, and pricing tables live in actor/README.md — that's the file rendered on the Actor's Apify Store page.

Use this repository if you want to:

• Read the source code for a working multi-platform Apify Actor (Crawlee + Camoufox + Apify Standby + MCP server).
• Fork and adapt the architecture for arbitrage in other reference-driven secondary markets (sneakers, art, fine wine, graded cards, handbags).
• Open an issue about a bug or a feature request.
• Contribute (see CONTRIBUTING.md).

Repository layout

.
├── actor/                   # The Apify Actor itself (deployed to Apify Cloud)
│   ├── .actor/              # Actor manifest, input/output/dataset/KV schemas, OpenAPI 3.0
│   ├── .github/workflows/   # Local actor CI (test on push)
│   ├── assets/screenshots/  # Images embedded in actor/README.md
│   ├── src/
│   │   ├── crawlers/        # One file per source platform
│   │   ├── utils/           # Brand detection, URL builders, FX
│   │   ├── aggregator.ts    # Sub-ref grouping + trimmed median + spread detection
│   │   ├── alerts.ts        # Telegram dispatcher (with 24h dedup)
│   │   ├── server.ts        # Apify Standby HTTP server (REST + MCP JSON-RPC)
│   │   └── main.ts          # Entry point (mode dispatcher: batch vs standby)
│   ├── test/                # Jest unit tests
│   ├── Dockerfile           # Container for Apify build pipeline
│   ├── README.md            # The Apify Store landing page (full product docs)
│   └── package.json         # apify, crawlee, camoufox, express
├── .github/workflows/       # Repo-level CI (build + test on every push)
├── LICENSE                  # MIT
└── CONTRIBUTING.md          # Contribution guidelines

Quick start (run locally)

git clone https://github.com/DataKazKN/watch-arbitrage-mcp.git
cd watch-arbitrage-mcp/actor
npm install
npm run build

# Local run (writes to ./storage/ — not synced to Apify Cloud)
npx apify run

# Or test the server entry point directly
node dist/server.js

To deploy your own fork to Apify Cloud:

npm install -g apify-cli
apify login
apify push

Per Apify's GitHub integration guide, you can also wire this repo to auto-rebuild your Actor on every git push via a webhook — see .github/workflows/apify-build.yml.

Architecture highlights

• Per-platform crawler files — every source has its own ~100-line module conforming to the same Listing shape, so the aggregator stays platform-agnostic.
• Camoufox + Apify proxy rotation for Cloudflare-protected sources (Chrono24, Bobs).
• Sub-reference grouping — listings are grouped by extracted sub-reference (e.g. 5711/1A-010 vs 7011/1G), not by user search term, to avoid conflating men's vs ladies' sub-models in the median computation.
• Trimmed median — drops the top/bottom 10% before computing the median, to absorb dealer-listing outliers (e.g. one $3.8M typo or one $200K above-market piece) without skewing the spread baseline.
• Dual entry point — same Docker image runs as a batch crawler under cron, OR as an MCP server in Apify Standby mode (handled by Actor.config.get('metaOrigin') === 'STANDBY' in main.ts).
• Pay-Per-Event monetization — charges per actor-start, per reference-monitored, per apify-default-dataset-item, and per spread-alert-triggered (the primary value event). Spending limit (ACTOR_MAX_TOTAL_CHARGE_USD) is honored, so a runaway alert spike never blows past the user's cap.

Documentation

• Full product README + dealer tutorial + pricing
• Input schema
• Output schema
• Dataset schema
• OpenAPI 3.0 spec for MCP server

License

MIT — fork freely, build commercial products, no attribution required.

Contributing

Issues and pull requests welcome. See CONTRIBUTING.md. Adding a new source platform takes roughly 30 minutes following the existing crawler template (actor/src/crawlers/chrono24.ts is the cleanest reference implementation).