Amazon reviews API for buyer review data

Amazon does not publish a general-purpose API for pulling review text off its own product pages, and the reviews shown to a logged-out visitor are only a small featured sample. Anyone who wants full, page-by-page review data has to work through the same review pager the website itself uses, which behaves differently for a signed-in shopper than it does for an anonymous one.

Roketfy's Amazon reviews API sits in front of that pager. You send an ASIN or product URL, and the endpoint returns clean, structured JSON — rating, title, text, date, author name and helpful-vote count — instead of a page of HTML you would otherwise have to parse yourself.

Every other endpoint in the Data Engine runs on infrastructure Roketfy manages end to end. Amazon reviews are the one exception: the pager behind them answers a signed-in browser session and nothing else, so the endpoint takes a cookies parameter — a session you supply, sent along with your request.

Amazon mints a CSRF token per session and refuses a review request that doesn't carry both that token and the cookies it was minted for. A single shared session across every Roketfy customer would mean one Amazon account rate-limiting everyone on the platform, and it would only ever surface the reviews that one account can see, in its own marketplace and language. Using your own session avoids both problems: you get reviews for the marketplace you're actually signed into, at your own pace.

The cookies you send go to Amazon and nowhere else — nothing on the Data Engine stores or echoes them back. If a session expires, or you sign out of the browser you copied it from, the endpoint reports the failure plainly (a missing CSRF token) rather than silently answering an empty page, so you know to copy a fresh cookie header and retry.

A page of Amazon reviews costs 10 credits and returns Amazon's own page size — the rating, title, review text, date, author and helpful-vote count for each review, plus the review count and seller the page header states. Amazon pages ten reviews at a time, so a listing with a few hundred reviews is a handful of pages, not a single call.

Only successful requests are ever charged. If a call fails — an expired cookie, a bad ASIN, an Amazon-side error — you don't pay for it, so there's no cost to retrying after you've refreshed your session.

For a product with a handful of pages, calling the reviews endpoint directly with an incrementing page number is enough. For a listing with hundreds of pages, the endpoint also supports Roketfy's batch task queue: you queue a task with your ASIN, cookies and a page cap, then poll it while it works through every page in the background, reading results as they land rather than waiting for the whole run to finish.

Batching bills the same 10 credits per page as a live call — a 200-page pull costs 2,000 credits regardless of whether you fetch it page by page or as one queued job. The advantage of the queue is operational: it survives a long-running pull without you having to keep a script open, and a page that fails costs nothing, exactly as it would live.

Yes. The starFilter parameter takes Amazon's own three options — all_stars, positive (3-5 stars) or critical (1-3 stars) — so you can pull just the reviews you actually care about instead of paging through everything and filtering client-side.

You can also target a specific Amazon marketplace with the domain parameter, from amazon.com and amazon.co.uk through amazon.de, amazon.com.tr and more than a dozen others — useful when you're tracking how the same ASIN reads to buyers in different countries.