Naver Scraper

Scrape public Naver shopping results by keyword and export product names, prices, stores, ratings, delivery details, images, and category identifiers to JSON, CSV, or Excel. This Naver shopping data collector works from the product records shown in ordinary public search results, making it useful for focused Korean product discovery and repeated price snapshots.

The actor accepts several Korean or English search terms in one run. It writes one row per distinct listing per keyword, with direct product links and the search term that found each listing. It is designed for the small shopping selection shown in public general search. A broad catalogue crawl needs many well-chosen terms because each term exposes a limited selection.

What data can you extract from Naver?

Each dataset row describes one product placement. Prices and delivery fees are amounts in Korean won; a blank value means the search listing did not provide it. The output keeps identifiers as strings so spreadsheet imports do not lose digits. The table below covers every output field.

Field	Description	Example
keyword	Search term that produced the row	러닝화
searchUrl	Public result page used for the row	https://search.naver.com/search.naver?query=...
searchPosition	Position within its shopping section, starting at one	1
sourceSection	Product section on the public page	shopping or nstore
isSponsored	Whether that placement is marked as sponsored	true
nvMid	Listing identifier used to remove duplicate placements	60704830910
channelProductId	Product identifier in the selling channel	5446
catalogId	Matched comparison catalog identifier, if supplied	null
productName	Product title without search highlight tags	러닝화 ...
productUrl	Direct desktop product URL	https://.../product/detail.html?...
mobileProductUrl	Direct mobile product URL, if supplied	http://m...
imageUrl	Primary product image	https://shopping-phinf.pstatic.net/...jpg
images	Image URL array from the listing	[..., ...]
mallName	Store name shown for the product	A store label
mallId	Store identifier, if supplied	gtsglobal
mallUrl	Store URL shown in the listing	https://...
salePrice	Positive regular price, if supplied	89000
discountedSalePrice	Discounted price, if supplied	79000
price	Selling price used for price filters	79000
discountedRatio	Discount percentage shown by the source	11
priceCurrency	Currency code	KRW
averageReviewScore	Average review rating	4.9
totalReviewCount	Total review count	179
averageReviewScore6Month	Recent average from the store section, when present	4.78
totalReviewCount6Month	Recent review count from the store section	7517
purchaseCount	Purchase count displayed with the listing	3094
mallCount	Number of selling stores when reported	0
deliveryFee	Displayed delivery fee in won	0
freeDelivery	Whether the displayed fee equals zero	true
fastDeliveryType	Source delivery-speed label	NONE
lCatId	Top-level category identifier	50000001
mCatId	Second-level category identifier	50000174
sCatId	Third-level category identifier	50001466
ssCatId	Fourth-level category identifier	50003854
isBrandStore	Source flag for a brand store	false
isSmartStoreProduct	Source flag for a store product	false
isHotDeal	Source hot-deal flag, if supplied	false
productType	Product type label from the source	NORMAL
collectedAt	UTC collection time for the run	2026-09-27T20:34:46+00:00

The store section sometimes adds a six-month review average and count that the shopping section does not include. The shopping section can contain both sponsored and ordinary placements of the same listing. The actor removes repeated nvMid values within each keyword while preserving the first eligible placement. The same product may appear once for each of two different keywords; those separate rows preserve search context and let you measure where the product appeared.

price prefers discountedSalePrice when available, then falls back to a positive salePrice. A listed regular price of zero is treated as unavailable rather than as a free product. The discount percentage is the source value and is not recomputed from the two price fields. Use the price and timestamp together when comparing runs because listings, promotions, and stock can change.

How to scrape Naver

1. Enter one or more shopping search terms in Shopping keywords. Korean terms usually reflect the names used in the local catalogue. Start with terms that describe different categories or product families.
2. Choose a total product limit if you need a fixed dataset size. Set a price range only when you want the actor to discard results outside that range.
3. Decide whether to include sponsored placements. Leaving them on reflects the complete public selection; turning them off keeps eligible ordinary placements.
4. Run the actor. It checks the public general search page for each term and records the shopping and store product collections shown there.
5. Open the dataset and download JSON, CSV, or Excel. Use keyword, nvMid, and collectedAt to compare later snapshots.

For a first run, use several unrelated terms rather than expecting a large result from one broad term. A focused page may show fewer than ten distinct products. More terms are the practical way to collect more products from this public surface.

Input

Parameter	Type	Default	Description
keywords	String array	Required	Independent shopping search terms; Korean is usually the most specific choice.
maxProductsPerKeyword	Integer	12	Maximum distinct products retained for each term. The public page may expose fewer.
maxItems	Integer	0	Hard cap across all terms; zero removes the total cap.
includeSponsored	Boolean	true	Keep sponsored placements along with ordinary ones.
minPrice	Integer	Unset	Minimum displayed selling price in won.
maxPrice	Integer	Unset	Maximum displayed selling price in won.
proxyConfiguration	Object	Enabled	Proxy settings for public search requests.

A realistic product research input is:

{
  "keywords": ["러닝화", "세탁세제", "선크림", "커피믹스"],
  "maxProductsPerKeyword": 12,
  "includeSponsored": true,
  "maxItems": 0
}

That input was used for the verification run and produced 35 rows. The counts for an individual term can change between runs. The maximum is an upper bound, never a promise that every term contains twelve available products. If you specify a price range, filtering happens after the public search page is read; it cannot make the source return additional candidates beyond the visible selection.

Output example

The following values come from a product returned by the verification run. The example shows a useful subset of its fields; the dataset row also includes the remaining fields from the table above.

{
  "keyword": "러닝화",
  "sourceSection": "shopping",
  "searchPosition": 1,
  "isSponsored": true,
  "nvMid": "60704830910",
  "channelProductId": "5446",
  "catalogId": null,
  "productName": "밸롭 글라이드 러닝화 오프화이트 코리아에디션",
  "productUrl": "https://ballop.co.kr/product/detail.html?product_no=5446&cate_no=1287&display_group=1&utm_source=naver&utm_medium=shopping&utm_campaign=knowledge_shopping&mkt_in=Y&ghost_mall_id=naver&ref=naver_open",
  "imageUrl": "https://shopping-phinf.pstatic.net/main_6070483/60704830910.jpg",
  "mallName": "밸롭본사공식몰",
  "salePrice": null,
  "discountedSalePrice": 79000,
  "price": 79000,
  "priceCurrency": "KRW",
  "averageReviewScore": 4.9,
  "totalReviewCount": 179,
  "deliveryFee": 0,
  "freeDelivery": true,
  "lCatId": "50000001",
  "mCatId": "50000174",
  "sCatId": "50001466",
  "ssCatId": "50003854"
}

The product URL is a direct selling-page link supplied by the search result; it may point to a store outside the main shopping domain. The image URL is useful for catalogue matching and visual review. For spreadsheet work, price is usually the best single price column, while salePrice and discountedSalePrice explain how that value was chosen.

Use cases

• Korean product discovery: Collect a manageable set of visible listings for several product categories before deeper manual research. The keyword column makes it clear which query surfaced each listing.
• Price snapshots: Schedule the same terms at a regular interval, then compare nvMid, price, and collectedAt to detect changes. Check the direct product page before acting on a large change because search prices can reflect temporary promotions.
• Store comparison: Group rows by mallId or mallName to see which stores appear for a family of terms. The output records the selling store shown for each listing, not a complete table of every merchant selling a catalog item.
• Assortment tracking: Compare product identifiers and categories between runs to see which listings enter or leave the public selection. A missing row can also reflect ranking changes, so do not equate absence with product removal.
• Review-led screening: Sort the visible products by averageReviewScore and totalReviewCount to find items worth inspecting. These are aggregate numbers shown with the product; individual reviews are not collected.
• Delivery screening: Filter returned rows for zero delivery fee or a source fast-delivery label. Delivery eligibility can depend on shopper location and checkout conditions, so treat listing labels as a search snapshot.

These workflows work best when the query list is stable. Reusing the same terms gives you a comparable sample. If you change terms, the row count and product mix may change for reasons unrelated to the products themselves.

How much does it cost to scrape Naver?

The current result charge is $0.001592 per product row, or $1.592 per 1,000 rows. Charges follow delivered dataset items. Searches that return no retained product rows do not add result charges, although normal platform resource usage and account conditions may still apply. Check the run screen for the final amount.

Delivered rows	Result charge
35	About $0.056
1,000	$1.592
5,000	$7.96

The 35-row example reflects the verified four-keyword input above. Reaching 1,000 or 5,000 rows takes many varied terms because the public page yields a small selection per term. A narrow price band, removal of sponsored placements, or repeated product appearances can lower the delivered count and its corresponding result charge.

Set maxItems when you need an upper bound on output volume. For example, maxItems: 100 stops after one hundred delivered rows even if the input list contains more terms. If a term returns fewer records than maxProductsPerKeyword, the actor moves to the next term. A run also stops adding rows when its charged-result budget is reached.

Tips for best results

Use terms that match how local shoppers describe products. Brand and model terms can be precise but may return fewer visible cards than a broad category. If the goal is a broader survey, split a category into meaningful terms such as product type, intended use, and size. Keep those terms in the dataset for later interpretation.

Start without a price filter to understand the visible result mix. Then add minPrice or maxPrice if the same term surfaces items outside the range you need. Prices are filtered against the chosen selling price in the output, after discounts where the listing provides one. The filter does not request a deeper price slice from the source. If you need midrange coverage, choose narrower terms describing that segment.

Use includeSponsored: false when ranking analysis should exclude placements explicitly marked as ads. The page may also show the same product in sponsored and ordinary positions. The actor keeps the eligible ordinary placement if it appears later. When sponsored listings are included, the first occurrence for a product is kept, so the isSponsored flag describes the retained placement.

Use nvMid for product-level matching and channelProductId when you need the identifier of a store listing. catalogId is often empty because the public card does not always expose a matched comparison catalog. Keep identifiers as strings in spreadsheets. Numeric import can round long values and make joins unreliable.

A new run fetches the current page for each keyword. Search rankings, promotions, delivery labels, review totals, and seller links can change. For monitoring, retain each run’s dataset or add its items to a dated table. Compare several snapshots before interpreting a short-lived difference.

Limits

This actor collects the public product records embedded in ordinary Naver search results. In research, broad queries exposed about 7–12 distinct products each after duplicate placements were removed. The dedicated shopping search and catalog pages returned an HTTP 418 access restriction, including through a Korean residential proxy. The shopping JSON endpoint exposed by the page returned an empty list in the tested anonymous session. This actor therefore does not claim full shopping pagination, 10,000 products from one keyword, complete seller tables, product specifications, review text, or price history.

The public page may omit a section for a term or change its embedded data structure. Some fields are available in only one section. For example, six-month review figures appeared with certain store results, while comparison catalog identifiers were sparse. A blank field is an honest absence from the listing rather than a guessed value. Category fields are source identifiers; names and full category paths are not provided by this extraction.

Product links can point to external stores, and their availability is outside the search page’s control. Search price and delivery information can differ from checkout values. This actor collects publicly visible listing information only and does not log in or access private shopper data. If an input keyword returns no product state, the run logs that term and continues with the next; an entirely empty run fails visibly.

The verified input returned 35 rows with product names, links, prices, store names, images, identifiers, and timestamps present on every row. That is a check of one live run, not a guarantee of future result counts. For coverage beyond this public selection, use a source that has authorized access to deeper shopping results.

Use with AI agents (MCP)

The input schema gives automated clients a small set of explicit parameters, and each dataset row has stable field names. Two useful requests are:

Search 러닝화, 세탁세제, and 선크림; collect up to twelve products per term; return product name, displayed price, store name, and direct URL.

Run the same four Korean shopping terms every morning; compare nvMid and price with yesterday’s rows; list changed prices with both collection timestamps.

Keep an explicit product limit in automated workflows so cost and output size remain predictable. Ask for keyword and collectedAt alongside prices when comparing runs, because a price without its search context or time can be misleading.

FAQ

Is the data public, and may I use it? The actor reads public search listing data without a login. Your use of the resulting data should comply with applicable terms, copyright rules, privacy law, and the rules of your own market. Avoid using listing links or store identifiers for unwanted contact.

Do I need to configure a proxy? The default input enables a proxy. Ordinary search results were available through the project’s remote network during verification. The dedicated shopping pages remained restricted, which is why the actor uses general search results.

How fast is a run? The verified four-keyword run finished within about a minute. Network retries, proxy changes, and the number of keywords can change runtime. The actor processes terms in order and stops once a total cap is reached.

Can I schedule and monitor the same searches? Yes. Save a fixed input and schedule repeat runs. Export each dataset with its collection time, then join rows by nvMid and keyword. A product disappearing from a shallow result selection does not prove it has been discontinued.

Can I export to a spreadsheet? Yes. Download CSV or Excel from the dataset. Use the overview columns for quick inspection and the full JSON rows for all optional fields. Import the ID fields as text to preserve their exact digits.

Why is a field empty? The public listing did not include it for that product or section. Some store results carry recent review figures, some products have no matched catalog ID, and some delivery details depend on the seller. The actor does not fill absent values with estimates.

Why did a keyword produce fewer than twelve rows? Twelve is the default stopping limit, while the public page often shows fewer distinct products. Sponsored and ordinary sections can repeat a listing. Price filters and includeSponsored: false can reduce the remaining rows further.

Integrations

Use the actor’s dataset through its API or connect a finished run to webhooks, scheduling, automation flows, or a spreadsheet import. A practical monitoring workflow stores the JSON rows with the run date, joins on nvMid and keyword, and reports price differences only where both runs have a positive displayed price. Keep the direct product URL in the exported table so a reviewer can check a listing before making a purchasing or merchandising decision.

Support

Open an issue on the Issues tab with the search term, run link, and the field you expected. We reply within 24 hours and add fields on request when they are available in the public result data.