OLX ships a 3.7 MB page built with Emotion, so the class names look like css-1tqlkj0 and rotate whenever the CSS changes. Selecting on those is how an olx scraper breaks silently a month after you write it.
The page also ships data-testid attributes on every element you actually want. Those exist for OLX's own test suite, which means their CI breaks when they move. That is the contract this repo builds on, and it returns 52 listings for 1 credit.
Built on ScrapingBee's web scraping API. Verified live on 2026-09-15 against olx.pl.
rules = {"listings": {"selector": 'div[data-testid="l-card"]', "type": "list", "output": {
"title": {"selector": 'a[data-testid="card-title-link"]', "output": "@aria-label"},
"url": {"selector": 'a[data-testid="card-title-link"]', "output": "@href"},
"price": 'p[data-testid="ad-price"]',
"location_date": 'p[data-testid="location-date"]',
}}}Live result, 1 credit:
{
"title": "Wymiana baterii iPhone 15 PRO MAX na poczekaniu Mińsk Mazowiecki",
"url": "/d/oferta/wymiana-baterii-iphone-15-pro-max-...-ID1an6LV.html?search_reason=search%7Cpromoted",
"price": "...",
"location_date": "..."
}52 cards on one page, which is roughly double what most marketplace listings return per request.
The visible anchor text is truncated with an ellipsis when a title is long. The aria-label on the same element carries the full untruncated title, because screen readers need it. One attribute change, complete data.
href comes back as /d/oferta/... with no host, so prefix the domain you queried. It also carries a search_reason query parameter, and search_reason=search|promoted is how you tell a paid placement from an organic one without a separate field:
from urllib.parse import urlsplit, parse_qs
parts = urlsplit(row["url"])
promoted = "promoted" in parse_qs(parts.query).get("search_reason", [""])[0]
clean = f"https://www.olx.pl{parts.path}"Strip the query before you deduplicate, or the same listing appears twice because it was reached two ways.
The tail of the slug is -ID<code>.html, for example ID1an6LV. That code is the stable key across repeated pulls. The rest of the slug is cosmetic and changes when a seller edits the title.
OLX is a family of national marketplaces, not one site. The verified page here is olx.pl, which is Poland, and the landing page for this target names olx.pl as the source too. Other countries live on their own domains, such as olx.ro, olx.bg, olx.ua and olx.kz, each with its own currency and language.
The search path shape on the Polish site is:
https://www.olx.pl/d/oferta/ one listing
https://www.olx.pl/d/oferty/q-<query>/ a search
Prices come back as display strings in local currency, complete with the local thousands separator and a zł suffix on the Polish site, so normalise per domain rather than with one global parser.
The request also reported spb-initial-status-code: 301, meaning OLX redirects to a canonical form. That is a redirect being followed, not a problem.
mode=auto walked the ladder and settled on the cheapest rung. The delivered HTML already contains all 52 cards, so there is nothing for a browser to add.
Two structured data blocks also come down with the page, a WebPage and a BreadcrumbList. The breadcrumb gives you the category path, which saves parsing it out of the URL.
Measured from spb-cost response headers:
| Call | Credits |
|---|---|
A search results page via mode=auto |
1 |
| Rejected request | 0 |
mode=auto bills only the rung that worked and nothing at all if every rung fails. It cannot be combined with render_js, premium_proxy or stealth_proxy, and sending both returns HTTP 400 while billing nothing.
At 52 listings per credit, tracking 20 search queries hourly is 480 credits a day. The entry paid tier of 250,000 credits covers that comfortably. ScrapingBee does not cache, so poll on the cadence your prices actually move.
Plan tiers are on the pricing page.
Public listing and search pages. Seller accounts, the messaging system, phone numbers revealed only after a click and anything requiring a signed in session are out of reach, and scraping under login credentials is prohibited by ScrapingBee's terms of service.
OLX is a consumer to consumer marketplace, so most sellers are private individuals rather than businesses. Listing text, location and any contact detail are personal data, and harvesting seller contact information is specifically not what this is for. Track prices, categories and market supply, and leave the people out of your dataset. OLX's Regulamin governs the Polish site, with equivalents per country domain.
Reference: extraction rules, data extraction feature, markdown output.
Adjacent marketplace endpoints: classified API, Craigslist API, Craigslist ad API, Gumtree API, Avito RU API, eBay scraper, marketplace API, Etsy API.
Why are my OLX selectors breaking?
Because you selected on an Emotion class such as css-1tqlkj0. Those are generated and rotate. Use the data-testid attributes, which OLX's own tests depend on.
Why is my title cut off with an ellipsis?
Because you read the anchor text. Read @aria-label on the same element for the full title.
How do I tell promoted listings from organic ones?
The search_reason query parameter on the listing URL. search|promoted marks a paid placement.
Which OLX domain should I use?
The one for the country you care about. OLX is a set of national marketplaces, and the verified example here is olx.pl.
What is the stable ID for a listing?
The ID<code> segment at the end of the URL path. The slug before it changes when a seller edits the title.
Do I need JavaScript rendering? No. All 52 cards are in the delivered HTML at 1 credit.
MIT. See LICENSE.