Hepsiburada Products Scraper

Product listings from Hepsiburada, Turkey's second-largest marketplace: price in lira, the instalment offer, rating and review count, badges, image and product URL — for any search term.

HTTP only, no browser, no key. Off Hepsiburada's own server-rendered listing.

Input

field	what it does
searchTerms	laptop, kulaklık, iphone 15 — one target per term.
maxItems	Overall cap. One term tops out at ~1,800 (see below).
maxConcurrency, minRequestInterval	Pacing. Pages are 3.5 MB each.

Four things about this data worth knowing

1. The first price on the card is the instalment, not the price

A card's text, in the order it appears, for a 32,999 TL laptop:

'Lenovo IdeaPad Slim 3 …', 'Ürün puanı 4.9, 23 değerlendirme', '4,9',
'Kampanya: Peşin fiyatına 6 x 5.499 TL', '6 x', '5.499 TL',   ← first "N TL"
'32.999', 'TL'                                                 ← the price

"Peşin fiyatına 6 x 5.499 TL" is six instalments at the cash price. A scraper that takes the first string containing "TL" reports 5,499 TL for a 32,999 TL laptop — wrong by about 6×, and nothing about 5,499 looks wrong for a laptop. This Actor reads the price from its own node only, and keeps the instalment beside it as installmentCount × installmentAmount.

2. Two id spaces on one page

…-83k700pstr-p-HBCV0000FC0KHB      a SKU   (one specific variant)
…-bilgisayar-pm-HBC0000D9H7T8      a MODEL (a product with variants)

Half the cards on a page are one, half the other. A parser keyed on -p- alone returns null ids for the model half, and a run that de-duplicates on that id then collapses them all into one row. Both are read; every row carries idType, and the summary counts skuRows and modelRows so a regression in either is visible.

3. Turkish numbers, with the decimals in a separate node

68.238 + ,40 + TL is 68,238.40 lira: . is the thousands separator, , the decimal mark, and the ,40 lives in a nested node whose text is just TL when there are no decimals. float("68.238") is wrong by 1000× and still looks like a plausible price for something. Every row keeps priceRaw so the arithmetic is auditable.

4. Page 50 answers HTTP 403 — and it is not a block

Page 49 returns 36 cards. Page 50 returns 403 with a 1,753-byte body; so do pages 500 and 9999, while page 1 keeps answering 200 in the same session. It is Hepsiburada's way of saying "no such page", but it reads exactly like a WAF kicking in mid-crawl. The wall is pinned at 49 and never requested — roughly 1,800 products per term — and reaching it is reported as ceilingHit.

Output

• PRODUCT — productId + idType, title, price + priceRaw + currency, installmentCount + installmentAmount + campaignText, rating, reviewCount, badges, imageUrl, productUrl.
• SEARCH_SUMMARY — productsReturned, matchCountAvailable: false, skuRows, modelRows, rowsWithoutId, duplicateRowsDropped, ceilingHit, stoppedReason.
• ERROR — one row naming what went wrong, instead of a silent empty.

Technical notes

• Every class name carries a build hash (price-module_finalPrice__LtjvY) that changes on deploy. Every selector matches on the stable stem with [class*=…], never on a full class name.
• One search in a handful comes back as HTTP 206 with the complete page. A 200-only classifier files that as fatal and loses the term; 2xx is treated as success here and the card count decides.
• Two kinds of empty page. A nonsense term returns zero cards with a "bulunamadı" marker — an honest no-results. An empty term returns zero cards without it — the search page with nothing searched. The empty term is refused locally; zero cards without the marker is reported as page_shape_changed rather than as "no results".
• No keyword-match check, deliberately. Titles are Turkish ("Taşınabilir Bilgisayar"), queries often English ("laptop"), and the failure such a check guards against elsewhere — a dropped keyword answered with an unrelated catalogue — was measured not to exist here. A check with no true-positive case and a routine false-positive case is worse than none.
• robots.txt checked 2026-09-16: no AI-bot group, no blanket disallow, /ara?q= allowed. The platform inventory was checked before a line was written.

Known limits

• The listing carries no struck-through price, no merchant name and no result total — checked on a discount-sorted search too. Product detail pages are not read.
• ~1,800 products per search term. Slice with narrower terms to go wider.
• Prices in lira only.