Amazon Search Scraper + Ratings Histogram

Drop-in compatible with junglee/free-amazon-product-scraper and igolaizola/amazon-search — 12× cheaper, and it doesn't fall over. Paste your existing input, change the actor ID, keep your field names.

Scrapes Amazon search results into one record per product — ASIN, title, price, list price, stars, ratings count, image, sponsored flag, position and page — and adds the thing that is genuinely hard to get anywhere else: the full 1–5 star ratings histogram on every product, plus the exact ratings count instead of the rounded "1.6K" the search page shows.

No login. No captcha-solver add-on. No residential proxy.

---

Why this one

The two actors this replaces are not slightly flaky — they are broken often enough that you notice. Numbers from the Apify Store, 19 September 2026:

Actor	users / 30 d	rating	failed runs / 30 d	price
junglee/Amazon-crawler	2 022	4.16★	8.4 %	$0.005 / result + extras
junglee/free-amazon-product-scraper	481	4.97★	17.4 %	$0.012 / result
igolaizola/amazon-search	71	3★	38.5 %	$0.001 / result
this actor	—	—	—	$0.001 / product

Roughly one run in six fails on the most popular "free" Amazon product scraper, and better than one in three on the cheap search one. If you are running these on a schedule, that is the difference between a dataset and a gap in your dataset.

What makes it reliable

• A real Chrome TLS fingerprint. Amazon fingerprints the TLS handshake, not just the IP. Plain curl/requests/httpx get a captcha from any IP tier; a genuine Chrome handshake gets the real page from an ordinary datacenter proxy. That is why this actor needs no captcha solver and no $8/GB residential traffic.
• Amazon's bot walls arrive with a 200. The "Click the button below to continue shopping" interstitial is a ~2 KB body served with a 2xx status. Scrapers that only check response.status_code treat it as success and finish "SUCCEEDED" with an empty dataset. This actor classifies any undersized 2xx body as a block, rotates to a fresh proxy session and cookie jar, and retries.
• Amazon A/B-tests its own markup. Every field is tried through several selectors and degrades to null rather than raising, so a class-name change costs you one field, not the run. A listing page that parses zero cards is retried once on a clean session before it is believed.
• Partial failures are reported, never hidden. The run status message says how many products came from how many start URLs, how many carry a histogram, how much traffic was used and how many sessions had to rotate. The run only fails if nothing was produced.

What you get that the incumbents don't

	this actor	free-amazon-product-scraper	igolaizola/amazon-search
1–5★ histogram	✅ ratingsHistogram + starsBreakdown	✅ starsBreakdown (requires the $0.012 detail fetch)	❌
Exact ratings count	✅	✅ (same)	❌ rounded "1.6K"
Price per product	$0.001	$0.012	$0.001
Search position / page	✅	✅ categoryPageData	❌
Multi-marketplace	✅ 21 domains, verified on .com and .co.uk	✅	via URL
Needs a captcha solver	❌	offered as an option	❌

The histogram is normally only on the 2.5 MB product page. This actor reads Amazon's own ~21 KB ratings-summary fragment instead — the same numbers, 1/120th of the bandwidth. That is the whole reason it can charge $0.001 where a detail-page scraper has to charge $0.012.

---

Input

Every field below is optional except that you must give something to search.

Field	Type	Default	Notes
categoryUrls	array of {url}	[{"url": "https://www.amazon.com/s?k=mechanical+keyboard"}]	Search, category or /dp/<ASIN> URLs. Same name and shape as junglee/free-amazon-product-scraper. Filters in the URL (rh=, s=price-asc-rank, department nodes) are preserved; only page is replaced.
searchTerms	array of strings	[]	Plain keywords, searched on domain.
search	string	—	A single search URL. igolaizola/amazon-search compatibility.
maxItems	integer	—	Alias of maxItemsPerStartUrl. igolaizola compatibility.
maxItemsPerStartUrl	integer	100	Products per start URL. 0 = everything Amazon returns.
maxSearchPagesPerStartUrl	integer	20	Listing pages per start URL (16 results per page).
domain	select	amazon.com	21 marketplaces. Start URLs keep their own marketplace, so a amazon.co.uk URL is never priced in USD.
scrapeRatingsHistogram	boolean	true	Adds ratingsHistogram, starsBreakdown and the exact ratingsTotal. Costs ~20 % more traffic.
proxyConfiguration	proxy	{"useApifyProxy": true}	Datacenter is enough for amazon.com — the Chrome fingerprint is what gets you in, not the IP tier.
residentialFallback	boolean	true	Non-US marketplaces wall off the datacenter tier, so a blocked marketplace is retried once on a residential IP in its own country, automatically. amazon.com never triggers it and never pays for it. Set false to pin a run to one tier.

Unknown fields from the incumbents (useCaptchaSolver, scrapeProductDetails, maxProductVariantsAsSeparateResults) are accepted and ignored, so a pasted input never errors.

How many results can I get?

Amazon serves 16 results per listing page and caps a search at 7–20 pages depending on the term, so one search term yields roughly 112–320 products. To go wider, split the term across category nodes rather than asking for more pages — the same limitation the incumbents document.

---

Output

One record per product. Fields in the first block carry junglee/free-amazon-product-scraper's exact names and shapes — including price as a {value, currency} object — so existing downstream code keeps working.

{
  "asin": "B08Z6X4NK3",
  "title": "Logitech G413 SE Full-Size Mechanical Gaming Keyboard - Black",
  "url": "https://www.amazon.com/dp/B08Z6X4NK3",
  "brand": null,
  "price":     { "value": 71.49, "currency": "$" },
  "listPrice": { "value": 89.99, "currency": "$" },
  "stars": 4.5,
  "starsBreakdown": { "5star": 0.77, "4star": 0.1, "3star": 0.06, "2star": 0.02, "1star": 0.05 },
  "reviewsCount": 1652,
  "thumbnailImage": "https://m.media-amazon.com/images/I/61thTSgqWOL._AC_UY218_.jpg",
  "isAmazonChoice": true,
  "amazonChoiceText": "Overall Pick",
  "inStock": null,
  "categoryPageData": {
    "categoryUrl": "https://www.amazon.com/s?k=mechanical+keyboard&page=1",
    "isSponsored": false,
    "productPosition": 1,
    "pageNumber": 1
  },

  "ratingsHistogram": { "5": 77, "4": 10, "3": 6, "2": 2, "1": 5 },
  "ratingsTotal": 1652,
  "reviewsCountIsApproximate": false,
  "rating": 4.5,
  "priceValue": 71.49,
  "listPriceValue": 89.99,
  "currency": "USD",
  "thumbnailUrl": "https://m.media-amazon.com/images/I/61thTSgqWOL._AC_UY218_.jpg",
  "isPrime": null,
  "isSponsored": false,
  "position": 1,
  "page": 1,
  "searchTerm": "mechanical keyboard",
  "domain": "amazon.com",
  "totalSearchResults": 37186
}

ratingsHistogram is percentages ({"5": 77} = 77 % of ratings are 5★); starsBreakdown is the identical data as fractions, in the shape the incumbent already emits. Use whichever your pipeline expects.

Honest nulls

A missing value is always null — never a guess and never a crash:

• brand — Amazon's current search card carries no brand element on most page variants. The line under the title is the feature blurb, not the brand, so it is not passed off as one.
• isPrime — logged-out search pages often show a "Join Prime" delivery upsell on every card, which distinguishes nothing. When the page variant has no real per-item Prime badge, this is null rather than a uniform false.
• inStock — only stated on product pages, so null for search results.
• reviewsCountIsApproximate — true means reviewsCount came from the card's rounded "1.6K". With scrapeRatingsHistogram on it is exact and this is false.

---

Switching from junglee/free-amazon-product-scraper

1. Change the actor ID to diopside/amazon-search.
2. That's it. categoryUrls, maxItemsPerStartUrl and maxSearchPagesPerStartUrl mean the same thing here, and title, url, asin, brand, price, listPrice, stars, starsBreakdown, reviewsCount, thumbnailImage, isAmazonChoice and categoryPageData come back with the same names and shapes.

What changes, deliberately:

• Cost drops from $0.012 to $0.001 per product. This actor reads search listings plus the tiny ratings fragment; it does not open each product's 2.5 MB detail page.
• Detail-only fields are not returned — description, features, attributes, bestsellerRanks, seller, offers, variantDetails and the rest of the 57-field detail record live on the product page. If you need those, you want a product-detail scraper, not this one. If you only ever used the listing fields, you were paying 12× for pages you threw away.
• currency is added as an ISO code ("USD"), because price.currency is the symbol "$" — which is also CAD, AUD and MXN.

Switching from igolaizola/amazon-search

Change the actor ID; search and maxItems work unchanged. You additionally get the ratings histogram, exact ratings counts, list prices, search position/page, Amazon's Choice badges and 21 marketplaces — at the same $0.001 per product, with a 38.5 % failure rate replaced by a proxy ladder that actually retries.

---

Marketplaces

amazon.com runs on Apify's datacenter proxy: the Chrome TLS fingerprint is what gets you in, not the IP tier.

Every non-US marketplace we measured (amazon.co.uk, amazon.de) refuses the whole datacenter tier — including exit IPs already inside GB and DE — with a ~2 KB "continue shopping" stub served on HTTP 200/202. Session rotation cannot fix a tier-wide wall, so the actor escalates that marketplace once to a residential IP in its own country and retries the start URL. You don't configure anything; the run's status message says escalated to RESIDENTIAL for amazon.co.uk when it happens.

Verified end-to-end on the platform: amazon.com (16/16 products with histograms, no escalation) and amazon.co.uk (10/10 with histograms, GBP prices, automatic escalation). The other 19 domains use the same code path but have not each been run.

---

Pricing

Pay per event:

Event	Price
Actor start	$0.00005
Product	$0.001

You are charged once per product record pushed to the dataset. 1 000 products ≈ $1.00. No platform-usage surcharge, no per-offer or per-ZIP-code extras.

Use cases

• Price and assortment monitoring — track a category's prices, list prices and ranking positions daily.
• Review-quality screening — the histogram separates a 4.3★ built on a healthy spread from a 4.3★ that is 70 % five-star and 20 % one-star. A single average hides that; this is the field most competitor research is missing.
• Keyword / share-of-search research — position, page, isSponsored and totalSearchResults per term show who owns a search term and how much of page one is paid.
• Product sourcing — filter a category by rating distribution and ratings volume before committing to a detail scrape.

Legality and scope

Public product listings only. This actor does not read reviewer names, reviewer profiles, review text, or anything behind an Amazon login — only the aggregate rating data Amazon publishes on the product card. Scraping publicly available product data such as prices and ratings is legal; see Apify's is web scraping legal?.

Support

Found a field that stopped parsing after an Amazon layout change? Open an issue on the actor — layout drift is treated as a bug and is what the daily canary run exists to catch.