For AI

Answers

How do I get Amazon reviews, product details and Best Sellers into Claude, Cursor or my own AI agent?

Short answer

Give the agent an API to call instead of a browser. okdata returns Amazon reviews (10 a page, newest or most helpful first), product details by ASIN and Best Sellers lists for amazon.com or another country store, over HTTP or one MCP URL that works in Claude and Cursor. A page of reviews costs $0.006; finding 5 competing products and reading the 100 newest and 100 most helpful reviews of each, with their details, costs $0.636. Amazon's own Creators API has no review data.

A founder weighing a portable espresso maker wants an agent to answer two questions: what do buyers of the five best sellers complain about, and which complaints are new this month? For that the agent needs three things from Amazon: which products sell, what each listing says, and the reviews. Opening amazon.com in a browser is shaky ground; developers building browser agents report that the full review list sends a logged-out visitor to a sign-in page. An API call returns the same data as JSON, wherever the agent runs.

Four endpoints

  • Amazon Best Sellers: a category's list in rank order, 50 a page; two pages cover the whole list. list also gives New Releases, Movers & Shakers, Most Wished For and Gift Ideas.
  • Amazon product search: 48 results a page as a shopper sees them, sorted by relevance, best sellers, most reviewed, newest or price.
  • Amazon product details: one ASIN with price, rating, number of ratings and photos. With view=full it adds the brand, the "About this item" bullets, the category path, Best Sellers Rank and units bought last month under ext.
  • Amazon product reviews: 10 reviews a page with stars, title and text, date, the variant bought, photos, helpful votes and whether the purchase is verified. Most helpful first by default; sort=recent gives newest first.

Every endpoint takes country: US (amazon.com) by default, or GB, DE, FR, JP, CA, MX, AU, IN and the other stores in the parameter list. Prices come in that store's currency.

The review job, priced

The cursor stops after 10 pages for each sort order, so one product gives up to 100 newest and up to 100 most helpful reviews; the two sets can overlap. For five competing products:

Job Calls Cost
Search the niche sorted by best sellers, keep the top 5 1 $0.006
Details of the 5 (Best Sellers Rank, bullets, brand) 5 $0.03
100 newest reviews of each 50 $0.30
100 most helpful reviews of each 50 $0.30
The whole job 106 $0.636
Any 1,000 reviews 100 $0.60

Prices are in US dollars, prepaid by card, Alipay or WeChat Pay, and a new key comes with $0.10 of credit. Each page is its own call and is charged even when it comes back empty, so stop when page.has_more is false. A request rejected before it reaches Amazon, such as one without an ASIN or with an empty balance, is free. The model that reads the reviews is billed by your model provider.

Code

import os, requests

API = "https://api.okdata.online"
H = {"Authorization": f"Bearer {os.environ['OKDATA_API_KEY']}"}

def call(path, **params):
    body = requests.get(API + path, headers=H, params=params, timeout=60).json()
    if body["code"] != "ok":
        raise RuntimeError(f"{body['code']}: {body['message']}")
    return body

def reviews(asin, sort="recent", country="US", pages=10):
    cursor, out = None, []
    for _ in range(pages):
        params = {"id": asin, "sort": sort, "country": country,
                  "fields": "rating,text,variant,published_at,verified_purchase"}
        if cursor:
            params["cursor"] = cursor
        body = call("/v1/amazon/products/reviews/list", **params)
        out += body["data"]["items"]
        if not body["page"]["has_more"]:
            break
        cursor = body["page"]["next_cursor"]
    return out

top = call("/v1/amazon/products/search", query="portable espresso maker", sort="best_sellers")["data"]["items"][:5]
low = [f"[{p['id']}] {r['variant']}: {r['text']}"
       for p in top for r in reviews(p["id"]) if (r["rating"] or 5) <= 3]
print(len(low), "low-star reviews; ask the model to group them into themes, with counts")

In Claude or Cursor, through MCP

The same endpoints are tools on the okdata MCP server at https://api.okdata.online/mcp. In Claude, on claude.ai or in the desktop app, add it as a custom connector and sign in; there is no key to copy (Claude setup). In Cursor, add it to ~/.cursor/mcp.json (Cursor setup):

{
  "mcpServers": {
    "okdata": {
      "url": "https://api.okdata.online/mcp",
      "headers": { "Authorization": "Bearer ${env:OKDATA_API_KEY}" }
    }
  }
}

Then ask in plain words:

  • "Find the five best-selling portable espresso makers on amazon.com, read the 100 newest reviews of each and group the one- to three-star complaints by theme, with counts and short quotes."
  • "Compare the About this item bullets and Best Sellers Rank of these three ASINs on amazon.de."
  • "What is on the Movers & Shakers list for kitchen on amazon.co.uk right now?"

find, describe, read_more and account are free; call is billed like the HTTP API. Ask the agent to state the cost before a long run and to cap the pages it fetches. For an agent of your own, wrap call as a tool; function calling from a Python agent shows the pattern.

Things that trip people up

  • Ratings are not reviews. The reviews number on a product is Amazon's count of ratings, not how many reviews you can page through.
  • Read low stars on their own. Filter to one to three stars before the model reads them, or praise drowns the complaints.
  • Variants tell you where it breaks. One listing often covers several colours or sizes, and variant says which one the buyer had.
  • Each store is its own market. Pass country=GB or DE to read that store's reviews, with prices in its currency.
  • Strangers wrote the text. Tell the model to treat reviews as data, never as instructions.
  • The Amazon endpoints are in preview. Fields may still change, and changes are announced beforehand.

What Amazon's own APIs give you

  • Creators API. Amazon has deprecated the Product Advertising API 5.0 and is replacing it with the Creators API, which is for Amazon Associates: you must be enrolled in the Associates program for the marketplace and have at least 10 qualifying sales in the past 30 days. Its resources (browse nodes, images, item info, offers, parent ASIN, search refinements, variation summaries) include no customer reviews.
  • Customer Feedback API. If you sell on Amazon, this part of the Selling Partner API gives review insights per ASIN or browse node: topics, month-on-month trends and customer snippets, in English, refreshed weekly, for the US, UK, France, Italy, Germany, Spain and Japan. It needs the Brand Analytics or Selling Partner Insights role. For your own listings it is worth setting up.

Other routes

For one or two products, reading the reviews yourself is often enough. Logged-in browser tools and open-source scrapers run on your own Amazon login and need fixing when Amazon changes its pages. Subscription research suites for Amazon sellers add keyword and sales estimates and are priced by the month. To turn the complaints into a product spec, see mining competitors' reviews; for the wider workflow, product research.

Sources

  1. Amazon Creators API introduction: Associates enrollment and at least 10 qualifying sales in the past 30 daysaffiliate-program.amazon.com · read
  2. Creators API resources: browse nodes, images, item info, offers, parent ASIN, search refinements, variation summaryaffiliate-program.amazon.com · read
  3. PA-API 5 deprecation notice: replaced by the Creators APIaffiliate-program.amazon.com · read
  4. Selling Partner API Customer Feedback API: review insights, roles, marketplaces, weekly refreshdeveloper-docs.amazon · read

Where people asked this (read 2026-11-17)

Updated

Connect once, then just ask.

Every new key comes with $0.10 to try it. One sentence connects your agent.

Start free
Connect your AI

Paste this into your agent

Read https://okdata.online/agents.md and set up okdata for me; my key is in the OKDATA_API_KEY environment variable.

Claude Code, Cursor, Codex and other agents that can open a link set themselves up from it.

MCP serverhttps://api.okdata.online/mcp