Live on RapidAPI · free tier, no credit card
It crawls a storefront the way a human would — category pages, pagination, product pages — and returns structured JSON: names, prices, photos and AI-written descriptions. No CSS selectors to configure, no per-store templates to maintain.
Illustration of a finished job: an extraction against www.heavyduty.com.br that crawled 160 pages, returned 50 products in 72 seconds with no duplicates, listing products such as “Broca Escalonada para Metal” in the category “Brocas”.
Numbers from one real production run against www.heavyduty.com.br — a site
with no /products path at all
How it works
Extraction is asynchronous on purpose: a real store takes minutes to crawl politely, and no HTTP request should be held open that long. You start a job, poll it, and collect the catalog.
POST /v1/extractSend the store URL. You get a job_id back immediately, plus the schema and limits the server actually applied. Optionally pass your own field list or a webhook.
GET /v1/jobs/{id}Poll for real progress, not a spinner: the current phase, pages visited, candidate pages found, pages extracted and items collected so far.
doneThe same endpoint returns the full catalog — every item, the pages that failed or were blocked by robots.txt, and the token usage the job cost.
# 1 — start the job
curl -X POST 'https://ai-product-catalog-extractor.p.rapidapi.com/v1/extract' \
-H 'X-RapidAPI-Key: YOUR_KEY' \
-H 'X-RapidAPI-Host: ai-product-catalog-extractor.p.rapidapi.com' \
-H 'content-type: application/json' \
-d '{ "url": "https://www.yourstore.com/" }'
# 202 Accepted
{
"job_id": "8f2c1a4e",
"status": "queued",
"poll": "/v1/jobs/8f2c1a4e"
}# 2 — poll while it crawls and reads
curl 'https://ai-product-catalog-extractor.p.rapidapi.com/v1/jobs/8f2c1a4e' \
-H 'X-RapidAPI-Key: YOUR_KEY' \
-H 'X-RapidAPI-Host: ai-product-catalog-extractor.p.rapidapi.com'
# 200 OK — real progress, not a spinner
{
"status": "running",
"progress": {
"fase": "extracao",
"paginas_visitadas": 160,
"paginas_candidatas": 126,
"paginas_extraidas": 41,
"itens_extraidos": 33
}
}# 3 — status "done": the whole catalog, in one payload
{
"status": "done",
"result": {
"itens": [
{
"nome": "Broca Escalonada para Metal",
"categoria": "Brocas",
"descricao": "Fabricada em aço rápido (HSS) com revestimento de titânio…",
"_url": "https://www.heavyduty.com.br/brocas/broca-escalonada-para-metal",
"_fotos": ["https://www.heavyduty.com.br/admin/files/produto/222/…jpeg"]
}
// + 49 more
],
"total_candidatos": 126,
"truncado": false,
"paginas_falhadas": [],
"paginas_bloqueadas_robots": [],
"usage": { "chamadas_ia": 60, "tokens_entrada": 113559, "tokens_saida": 7815 }
}
}Real payload shape, from the run above. Descriptions come back in the store's own language.
Why it works
Crawling a store is the easy half. Telling a real product page from a listing card, and a category from a product, is where scrapers quietly return garbage. That is the part this engine was built around.
Nothing to configure per store, and nothing to repair when a theme changes. The AI reads each page as text and decides what it is looking at — so a store you have never seen works on the first call.
Listing-card detection, deduplication by name and URL, and category pruning that drops the parent page when its child is the real product. Every one of those guards exists because a production catalog once came back wrong without it.
Send a field list — texto, number, bool, foto — and the AI extracts exactly those. Ask for a field called descricao and it writes one instead of copying the store's boilerplate. Presets ship for products and vehicles.
robots.txt is honoured, requests to the same host are rate-limited and serialised, the user agent says who we are, and every blocked page is reported back to you instead of silently skipped.
Jobs survive restarts, report incremental progress, and never hang: an orphaned job fails with a reason instead of pretending to run. Poll it, or hand us a webhook URL and forget about it.
WooCommerce, bespoke PHP shops, twenty-year-old catalogues with no /products path. When a store publishes schema.org product data we read that too — and when it does not, the page text is enough.
Where it stops
You are going to find these out on your first call anyway, so here they are before you spend one. Everything else about your store is fair game.
If the products only appear after JavaScript runs, we see an empty shell and return few or no items. The crawler reads the HTML the server sends.
A rendering engine for those stores is the next premium tier. Until it ships, a SPA storefront is a no.
The engine assumes each product has a page of its own. Catalogues where the SKUs live only inside a filterable listing — hundreds of variants behind dropdowns, no individual URLs — come back nearly empty.
That is the honest answer, not a failure: we tested exactly that case and the AI correctly reported “this is a listing page”, rather than inventing rows.
What people build with it
Move a catalogue between platforms without anyone retyping 400 products. Point at the old store, import the JSON into the new one.
Let a new seller paste their website instead of filling a spreadsheet. Their catalogue is live on your platform before they lose interest.
Re-run the same URL on a schedule and diff the results to watch assortment and pricing move, without maintaining one scraper per rival.
Feed a chat catalogue or an AI sales agent with real products, photos and descriptions — pulled from the store the merchant already has.
Pricing
Every plan runs the same engine, with nothing held back for a higher tier — the only thing that changes is how many calls it includes. Plans, billing and keys are handled by RapidAPI.
Enough to run your real store through it and judge the output.
Free
50 calls / month
No credit card
Start freeFor a migration project or a steady onboarding flow.
$ 19 /mo
400 calls / month
≈ $0.05 per call
Subscribe on RapidAPIFor a product that extracts catalogues every day.
$ 59 /mo
1,600 calls / month
≈ $0.04 per call
Subscribe on RapidAPIFor marketplaces and monitoring at volume.
$ 199 /mo
8,000 calls / month
≈ $0.025 per call
Subscribe on RapidAPICancel any time from your RapidAPI dashboard · a failed job returns a reason, not an invoice · need more than Mega? Ask us
FAQ
The run shown at the top of this page — 160 pages crawled, 50 products returned — finished in 72 seconds. A larger job typically takes 3 to 10 minutes, and crawling dominates that time, not the AI: we serialise requests to the same host instead of hammering the store.
Pointing at a category or listing page rather than the homepage is the fastest way to a result, because there is less of the site to walk.
Anything that serves its product pages as HTML: WooCommerce, custom-built shops, old catalogue sites, stores with no /products path at all. There is no per-platform integration to pick, because there is no per-platform code.
Two exceptions, both listed on this page: storefronts that render their grid in the browser, and catalogues where individual SKUs have no page of their own.
Yes, by default and in every phase — crawl, sitemap and extraction. We identify ourselves with a real user agent, keep a minimum delay between requests to the same host, and report the pages robots.txt kept us out of in the job result.
If you own the store, or hold written permission to extract it, a request flag lets you override that — and the declaration is recorded on the job.
Your plan's quota is counted in API calls, the way RapidAPI meters every request: the POST that starts a job, plus each poll you make while it runs.
In practice that keeps one extraction in the single digits — poll every 20 to 30 seconds rather than every second, or pass a webhook URL and spend exactly one call per catalogue.
Yes. Send a schema array of fields with a key and a type — texto, number, bool or foto — and the AI returns exactly those keys, whether you are extracting shoes, machine parts or used cars.
Naming a field descricao switches the model from copying the page to writing a description of its own. If you would rather not think about it, the produto preset covers name, brand, category, price, SKU, photos and description.
Fifty calls a month, free, no card. That is enough to run your real catalogue through it and decide with output in front of you instead of a promise.
50 calls/month on Basic · upgrade or cancel from your RapidAPI dashboard