Catawiki Publishes Its Expert Estimates in the Page JSON. I Built a Deal Finder on Top

Catawiki is an auction site where every lot gets reviewed by a category expert who writes an estimated value range. That estimate is the single most useful number on the page, and it turns out Catawiki ships it to your browser in structured JSON, for every lot, before you even bid.

Compare the current bid against the expert estimate across a whole category and you have a shopping list of undervalued lots. I built a bot that does exactly that. Here is what I learned about the site on the way.

The data is in NEXT_DATA

Catawiki is a Next.js app, so each page embeds its state in a <script id="__NEXT_DATA__"> block. No hidden API, no reverse engineering, just parse the JSON out of the HTML.

Two levels matter:

The search page carries pageProps.searchLots.lots, 24 lots per page. Handy for discovery, but there is a catch: search results do not include prices or estimates. Just IDs, titles and images.

The lot page is where everything lives: current bid, the expert's low and high estimate, whether the reserve price has been met, bid history, closing time. One fetch per lot, so a category scan is one search request plus N lot requests. I run them at concurrency 4 and the site has never complained.

The pivot I didn't plan

My original idea was to analyze completed auctions and build price history. Dead end: Catawiki's public search only exposes live lots. Once a lot closes it disappears from search, and there is no public archive to walk.

So the product pivoted from "price history" to "live deal finder": scan open lots, compute how far the current bid sits below the expert estimate, and surface the gap. The site constraint decided the product, which happens more often than product plans admit.

Fresh auctions look absurd, and that's fine

First real run: Game Boy lots showing bids 85 to 99 percent below the expert estimate. Too good to be true? Sort of. Auctions that just opened have starting bids of one euro against estimates of hundreds. The gap is real but everyone bidding knows it will close.

The actionable filter is time: a lot ending within 24 hours with the bid still 40 percent under estimate is a genuinely interesting signal. A lot with six days left is noise. So the bot takes endingWithinHours as its main input, and that single filter separates a gimmick from a tool.

One practical note: fetching through catawiki.com/es/ pins all amounts to EUR, which saves you currency detection entirely.

Try it

Packaged as an Apify Actor, catawiki-deal-finder:

curl -s -X POST \
  "https://api.apify.com/v2/acts/jdepablos~catawiki-deal-finder/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"query": "game boy", "endingWithinHours": 24, "minDiscountPct": 30}'

You get each lot with its current bid, the expert estimate range, the discount percentage, reserve status and closing time. Pricing is per auction scan, and a scan that returns nothing is never charged.

For AI agents there is an MCP route too:

https://mcp.apify.com?tools=jdepablos/catawiki-deal-finder

"Any watches closing today at half the expert estimate?" becomes a tool call your agent can run every morning.

Takeaways

Next.js sites are gifts: the state you want is usually sitting in __NEXT_DATA__, typed and complete, no DOM parsing needed. But check what the public surface actually exposes before you promise a product. I planned an archive analysis and the archive did not exist. The expert estimate data was better anyway: it is an opinion you can trade against, published in JSON.

If you hunt auctions and want a filter or a category I have not covered, comments are open.