Coupang Scraper

Collect public Coupang bestseller products and export product data to CSV or JSON. The verified output includes product identifiers, prices, ratings, delivery markers, bestseller positions, and source links.

Current status: The bestseller mode passed its live check on 28 September 2026 with 30 complete product rows. Keyword search remains blocked by an access challenge and has no verified output. Select browseBestsellers for the working mode; do not treat bestseller positions as keyword search rankings.

What data can you extract from Coupang?

Each verified row represents one product card on the public bestseller page. Field names are stable camelCase keys. Null means a value was not visible in that card. Boolean flags describe markers detected by the parser; a false value should be checked against the linked page when it matters.

Field	Description	Example or format
productId	Identifier in the product link or listing attribute	Numeric string
itemId	Item identifier from a listing attribute or link	Numeric string or null
vendorItemId	Vendor item identifier from a listing attribute or link	Numeric string or null
name	Product title printed on the listing	Korean or English text
url	Direct public product link	HTTPS URL
image	Product image from the listing	HTTPS URL or null
price	Displayed sale price	Integer Korean won
currency	Price currency	KRW
originalPrice	Higher reference price when displayed	Integer or null
discountRate	Displayed percentage discount	Integer or null
discountLabel	Text describing the discount	Text or null
unitPrice	Price per weight, volume, or item when shown	Text or null
coupon	Coupon message on the card	Text or null
cashbackText	Cashback or reward message	Text or null
ratingAverage	Displayed average star rating	Number or null
ratingCount	Displayed rating count	Integer or null
rocketDelivery	Whether a Rocket marker is visible	Boolean
deliveryBadges	Badge text found on the listing	Array of strings
deliveryPromise	Delivery timing text	Text or null
shippingText	Shipping message	Text or null
shippingFeeText	Shipping fee message	Text or null
freeShippingOver	Displayed threshold for free shipping	Integer or null
soldOut	Whether a sold-out marker is visible	Boolean
soldOutReason	Sold-out explanation	Text or null
lowStockText	Low-stock warning	Text or null
usedOffersText	Used-offer message	Text or null
sponsored	Whether the card is labeled as an advertisement	Boolean
searchRank	Position within a keyword search page; null in bestseller mode	Integer or null
bestSellerRank	Position among bestseller page product cards	Integer or null
couPick	Whether a recommendation marker is visible	Boolean
matchedKeywords	Search words found in the product title	Array of strings
badges	Distinct visible badge messages	Array of strings
keyword	Search term that produced the card; null in bestseller mode	String or null
searchPage	Search page number; null in bestseller mode	Integer or null
sourceMode	Origin of this row	bestsellers or search

A listing card does not expose a complete product detail record. Values such as seller identity, specifications, stock count, quantity-specific offers, and customer review text are outside this actor's current output. Prices may depend on coupon eligibility, membership, region, or time. A row is a snapshot of the displayed bestseller card, not a guarantee that a checkout will show the same total. When a downstream process needs a specific offer, follow the source URL and check it at the time of use.

The parser retains both identifiers and the URL because the URL can include an item and vendor item selection. These values matter when one product page represents several offers. The source mode and rank identify where each row was encountered. Bestseller mode removes duplicate product, item, and vendor item combinations within the page.

How to scrape Coupang

1. Enable browseBestsellers to collect the verified public bestseller page.
2. Set maxItems to the number of rows needed. The validated page supplied 30 product cards.
3. Optionally restrict rows by displayed price, rating, or Rocket delivery marker.
4. Run the actor, review the default dataset, and export CSV or JSON. A challenged or unrecognized page causes an error instead of an empty successful result.

The actor reads bestseller cards in page order and stops at the total row cap. A filter is applied after a card is read, so a restrictive filter may yield fewer rows than the maximum even when the page contains many products. The product limit is a ceiling, not a guarantee of coverage. Keyword mode remains available in the input for future access recovery, but it did not pass live validation.

Input

Name	Type	Default	Description
keywords	string array	Prefilled sample	Search phrases for keyword mode. Ignored when bestseller mode is enabled; keyword mode is currently blocked.
browseBestsellers	boolean	false	Read the public bestseller page and ignore any supplied keywords. This is the verified collection mode.
maxProductsPerKeyword	integer	100	Maximum retained products from each keyword. Zero removes this per-keyword cap, subject to the page and total caps.
maxItems	integer	0	Maximum rows across the run. Zero removes the total cap.
maxPages	integer	10	Maximum search pages requested for each keyword, from one to fifty.
sortBy	string	relevance	Relevance, best selling, low price first, or high price first.
minPrice	integer	Unset	Retain prices at or above this amount in Korean won.
maxPrice	integer	Unset	Retain prices at or below this amount in Korean won.
minRating	number	Unset	Retain rows with a displayed rating at or above this value. Unrated rows are excluded when this filter is used.
rocketDeliveryOnly	boolean	false	Retain rows marked for Rocket delivery.
proxyConfiguration	object	Enabled	Network configuration for public page requests.

The successful golden input collects the first 30 bestseller cards:

{
  "browseBestsellers": true,
  "maxItems": 30
}

The check requires at least 30 rows, with product ID, title, URL, price, currency, bestseller position, and source mode on every row. The 28 September 2026 run returned 30 rows and met every required fill threshold. Keyword search was tested separately and returned no usable rows.

A rating threshold applies to the number displayed on a listing card. It does not inspect individual reviews. The price filters use the current listed sale price, not the original price, coupon-adjusted checkout price, unit price, or any membership-specific amount. The Rocket filter uses markers visible on the card. None of these filters can recover products hidden behind page limits or an access challenge.

Output example

This is a real row from the successful 30-item bestseller dataset. Optional fields omitted here remain present as null or empty arrays in the full JSON export.

{
  "productId": "5385068615",
  "itemId": "19224496270",
  "vendorItemId": "86341163625",
  "name": "국내산 방아잎, 100g, 3개",
  "url": "https://www.coupang.com/vp/products/5385068615?itemId=19224496270&vendorItemId=86341163625",
  "price": 16840,
  "currency": "KRW",
  "cashbackText": "최대 842원 적립",
  "ratingAverage": 4.5,
  "ratingCount": 6484,
  "rocketDelivery": true,
  "deliveryPromise": "오늘(월) 도착",
  "soldOut": false,
  "sponsored": false,
  "searchRank": null,
  "bestSellerRank": 14,
  "keyword": null,
  "searchPage": null,
  "sourceMode": "bestsellers"
}

The example illustrates the difference between bestseller placement and keyword placement. The 14 is the card's position among the product cards collected from this bestseller page. It must not be interpreted as a rank for any search term. A title and price are present on this card, while discount, coupon, and shipping fields may be absent on another card.

Use cases

• Catalogue teams can compare products visible on the bestseller page between scheduled snapshots.
• Merchandising teams can review changes in the order of public bestseller cards.
• Delivery analysts can compare the share of listed products marked for Rocket delivery.
• Pricing analysts can inspect sale prices alongside visible reference prices and discount messages.
• Researchers can monitor ratings and rating counts for the products surfaced by the page.
• Inventory monitors can use visible sold-out and low-stock messages as signals to check a product page manually.

For repeatable comparisons, save the exact input and collection date with each export. Bestseller position can move between runs and may differ by location or session. Compare products using identifiers rather than title alone, because titles can change or repeat. Keep the URL so a reviewer can inspect the public listing behind a row. Treat absent optional fields as missing observations, not proof that an offer does not exist.

How much does it cost to scrape Coupang?

The configured result price is $2.392 per 1,000 rows. Ten rows cost about $0.0239 in result charges, and the verified 30-row bestseller run corresponds to $0.0718 in result charges. A total of 100 rows across several runs would be about $0.2392 in result charges. These examples use the configured per-row rate; they do not promise 100 distinct results from one bestseller page. The actor is not publicly released.

The total row cap controls how many items are pushed. It is not a spending guarantee for requests or browser retries. A restrictive filter can reduce the retained result count, while the same page still has to be fetched. Run-level platform usage can apply in addition to result charges. Inspect the actual run and dataset before projecting the cost of repeated collections.

Tips for best results

Enable browseBestsellers for the verified path. This mode ignores any supplied keywords. The bestseller page is a public selection, not a full catalogue. Save each dataset with its run date, then compare product IDs, prices, ratings, and positions with earlier snapshots. A product that disappears from this page has not necessarily been removed from the marketplace.

Start without filters. After inspecting the 30-row sample, add price or delivery restrictions that reflect a real question. A minimum rating can remove unrated items; it should not be used if those items are important to the study. A price limit is evaluated against the visible sale price, so a coupon or different quantity can change the effective checkout price without changing the filter result.

If a keyword run fails with an access challenge, changing the keyword is unlikely to fix it. The blocker was observed on search, product, and mobile routes. The public bestseller page was reachable in the successful run, but later access is not guaranteed. The actor rotates sessions and stops with a clear error when usable product cards do not arrive.

Limits

Public keyword search and product pages presented an access challenge, including with a residential session and browser attempt. Public sitemaps were reachable, but they contain links and update dates, not the price, rating, delivery, and ranking fields required for search. No live keyword search card could be inspected. The keyword parser is provisional. The bestseller parser was verified against a 30-row dataset from the public page.

The working mode covers the public bestseller page. It does not provide whole-category browsing, brand ID filters, item-condition filters, detail enrichment, or customer review extraction. The verified page yielded 30 cards; raising maxItems cannot manufacture additional rows. bestSellerRank records card order on that page, not a keyword rank or a claim about all categories. Displayed prices and availability can change between request and export.

The actor can fail before producing any dataset items when the page is challenged. Keyword mode currently has that behavior. A failure is preferable to a successful run with an empty dataset, because an empty dataset could be mistaken for a genuine no-results search. Downstream scheduled jobs should treat a failed run as missing data and preserve the last successful snapshot separately.

Automated workflows

Save the input with the exported dataset and run status. A scheduled process can request bestseller mode, check that the run succeeded, then compare IDs and prices with the previous successful dataset. Another process can flag products whose displayed delivery marker or page position changed. Both workflows require a successful source response; blocked runs must not be used as a market snapshot.

An automated consumer should check row count as well as run status. The golden check requires at least 30 bestseller rows and complete identity, price, rank, and source fields. Optional fields can remain null. Treat any unexpected drop in row count as a collection-health issue until the public page and source links are reviewed. Preserve the raw input so the same collection can be repeated.

FAQ

Is the actor working now? Bestseller mode passed a live 30-row check. Keyword search remains blocked and is not a verified collection mode.

Does it require an account? The verified source is a public, logged-out bestseller page. No account credentials are part of the input. Public visibility does not guarantee every later request will succeed.

Can I change the network settings? The input exposes network configuration. The verified bestseller run used the default path. Earlier search attempts tried multiple sessions and a residential route without obtaining usable search cards.

Why is a field empty? A bestseller card may not display that value. Null should be read as unavailable in the observed card, not as zero, false, or proof that the product has no such offer. Search rank and keyword are null by design in bestseller mode.

Can I collect detailed specifications or reviews? This version reads listing cards. It does not collect specifications, option matrices, seller details, or individual reviews.

Can I schedule and export results? Successful bestseller runs can be scheduled and their datasets exported in common formats. Monitor run status and row count because page access can change.

Can I use the data legally? Collection should be limited to public data and conducted in line with site terms and applicable law. Avoid misuse of personal data. This actor is intended for product listings rather than private account information.

Integrations

A successful dataset can be consumed through a dataset endpoint, scheduled collection, notifications, or a CSV export. Keep product IDs, URLs, bestseller positions, and collection timestamps in any downstream table. A workflow should alert on a failed run and on a count below its expected range. The verified dataset shows 30 rows; inspect source links when a decision depends on a specific offer.

Support

Open an issue on the Issues tab with the search input, run status, and the page or field that failed. Do not include account credentials or private checkout information.