Amazon MCP: Product, Price and Review Data for AI Agents

An Amazon MCP server is a Model Context Protocol server that gives an AI agent such as Claude, Cursor or ChatGPT tools for working with Amazon, but the phrase covers very different things: Amazon's own servers manage ad campaigns or AWS accounts, while a data server reads products, prices and reviews. For product data, the best Amazon MCP setup is Scrape.ly's hosted server, because the agent runs ready-made Amazon scrapers in the cloud against public pages, with no seller account, no Ads API credentials and nothing to install.

This guide sorts out what each "Amazon MCP" is for, connects the Scrape.ly server to each client, lists the Amazon Actors an agent can call, and shows two real runs from 26 October 2026: five mechanical keyboards from a search in 15.7 seconds for $0.0057, and the latest reviews of the top one in 7.9 seconds. It ends with the limits worth knowing before you build a price or review workflow on top.

Three things people mean by "Amazon MCP"

Google's first page for "amazon mcp" on 26 October 2026 mixes three kinds of server, and only one of them was ever about product data.

• Amazon Ads MCP Server. Announced in open beta on 2 February 2026, it "connects AI agents to Amazon Ads API functionality": creating and updating campaigns, reporting, account settings and billing data. It's "available globally in open beta to Amazon Ads partners with active API credentials".
• AWS MCP Server. A managed server that gives coding agents access to AWS services. Agents search AWS documentation without signing in and "authenticate through your existing IAM credentials" to make API calls. It's about cloud infrastructure, not the store.
• Shopping servers. The second organic result, Fewsats/amazon-mcp, let Claude search and buy Amazon products through a Fewsats payment account. Its README now opens with "This project has been discontinued and is no longer maintained."

None of these answers "what does this product cost today, how is it rated, and what are buyers saying?". That's the job of a data server.

How to connect the Scrape.ly MCP server

The Scrape.ly MCP server runs at https://mcp.scrape.ly on MCP's streamable HTTP transport, with OAuth sign-in or a Scrape.ly API token. A free account includes $5 of usage every month, no card required, which pays for about 4,700 product rows.

Claude Code. One command adds it. Inside a session, run /mcp and pick scrapely to sign in:

claude mcp add --transport http scrapely https://mcp.scrape.ly

Claude Desktop. Under Customize → Connectors, add a custom connector pointing at https://mcp.scrape.ly. Claude handles the OAuth sign-in.

Cursor. Use ~/.cursor/mcp.json for all projects or .cursor/mcp.json for one:

{
  "mcpServers": {
    "scrapely": {
      "url": "https://mcp.scrape.ly",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" }
    }
  }
}

Leave headers out to sign in with OAuth. Tokens are created under Settings → API & Integrations.

ChatGPT, VS Code, Windsurf and others. Any client with remote MCP support over streamable HTTP uses the same URL, with OAuth or a Bearer header. Scrape.ly's MCP docs have the VS Code config and a local stdio server.

To expose only the Amazon tools, list the Actors in the URL:

https://mcp.scrape.ly?actors=shelfbright/amazon-product-scraper,shelfbright/amazon-reviews-scraper,shelfbright/keepa-scraper

The Amazon Actors your agent can call

There's no single "Amazon tool". The agent picks from five Amazon Actors in the Scrape.ly Store, all maintained by Scrape.ly and all reading public pages without a login:

Prices are pay per event with platform usage included, as listed on each Store page on 26 October 2026. The Product Scraper adds $0.0004 per run and charges its extras only when you switch them on: $0.00151 per extra seller offer, $0.00151 per seller rating and $0.06003 per product priced for a delivery ZIP. The Amazon scrapers page compares them with sample output.

The four tool calls

Every Amazon request follows the same route through the server:

1. search-actors with "amazon" lists the Amazon Actors with prices and input fields.
2. fetch-actor-details returns the chosen Actor's schema and README. That's how the agent learns the Product Scraper wants categoryOrProductUrls and the Reviews Scraper wants productUrls.
3. call-actor starts the run with a JSON input and waits up to 45 seconds for it. Fast runs return finished; longer ones are followed with get-actor-run.
4. get-actor-output reads the rows page by page, with only the fields the agent names.

Rows are never pasted into the call-actor reply. A product row carries dozens of fields, including image galleries and A+ content, so reading asin,price,stars from a separate call keeps the context small.

A real run: comparing mechanical keyboards from a search

The prompt: "Use scrapely to search Amazon for mechanical keyboards, take the top five results, and compare price, rating and how many sold last month."

The agent sends the Product Scraper a search URL and a cap:

{
  "actor": "shelfbright/amazon-product-scraper",
  "input": {
    "categoryOrProductUrls": [{ "url": "https://www.amazon.com/s?k=mechanical+keyboard" }],
    "maxItemsPerStartUrl": 5,
    "maxSearchPagesPerStartUrl": 1,
    "scrapeProductDetails": true
  },
  "waitSecs": 45,
  "callOptions": { "maxTotalChargeUsd": 0.05 }
}

The run opened the search page and each product page and finished in 15.7 seconds: "Finished 1 URL(s): 5 products saved." The agent read asin,brand,title,price,listPrice,stars,reviewsCount,monthlyPurchaseVolume:

Four of the five sold below their list price, and the AULA board was Amazon's Choice and ranked #1 in PC Gaming Keyboards. One detail a careful agent catches: the cheapest result's title says "Mechanical Feeling Keyboard", which describes how it types, not the switches inside, so it shouldn't count in a mechanical-keyboard comparison. Every page showed "Deliver to Chicago 60601", the location Amazon picked for the request; prices can differ for other delivery addresses.

Cost: 5 products × $0.00106 + $0.0004 per run = $0.0057 at list price, platform usage included.

A real run: what recent buyers say about the AULA F75 Pro

The follow-up prompt: "Get the five most recent reviews of the AULA keyboard and tell me if anyone mentions a downside."

{
  "actor": "shelfbright/amazon-reviews-scraper",
  "input": {
    "productUrls": [{ "url": "https://www.amazon.com/dp/B0D14N2QZF" }],
    "maxReviews": 5,
    "sort": "recent",
    "filterByRatings": ["allStars"],
    "includeGdprSensitive": false
  },
  "waitSecs": 45,
  "callOptions": { "maxTotalChargeUsd": 0.05 }
}

It finished in 7.9 seconds over datacenter proxies only. Reviewer names are off by default, and we show reviewers as [reviewer n]:

All five gave five stars, so the agent dug into the text for downsides. One reviewer wished the knob clicked as it turns; another hadn't realized the letters are printed on the front of the keycaps rather than the top, which the listing title does say ("Side Printed PBT Keycaps"). A third wrote after more than a year of daily use and recommended the 2.4 GHz dongle over Bluetooth for lower latency.

For the overall picture the agent went back to the product row: 82% five-star, 9% four-star, 3% three-star, 1% two-star and 5% one-star across 1,877 ratings. That's the better source for a distribution, because public product pages show only a featured set of reviews.

Cost: 5 × $0.00439 = $0.02195 at list price. Both runs came to $0.02765 at list price; our account was billed $0.02212 after its plan's Store discount.

Prompts that work well

Name the Actor, the size and a spending cap:

• "Use scrapely to get these 20 ASINs every morning and tell me which ones dropped more than 10% below their list price."
• "Run shelfbright/keepa-scraper on B0D14N2QZF and tell me whether today's price is near its 90-day low."
• "Get the top 10 of Amazon's Best Sellers in Electronics with shelfbright/amazon-bestsellers-scraper and add each product's rating."
• "Pull the critical reviews of these three blenders with the Amazon Reviews Scraper and group the complaints."
• "Cap every run at $0.20."

The agent sets the last one as callOptions.maxTotalChargeUsd, and the run stops when it reaches the cap. For a full price-monitoring setup, see our guide to price scraping.

What to watch out for with Amazon data

• Reviews are the featured set. Signed-out product pages show a featured selection of reviews, about 8 in the Reviews Scraper's tests, and the full history sits behind an Amazon sign-in. Sort and star filters work within that set. Use the product row's star breakdown for distributions.
• Search stops at about 20 pages. That's roughly 200 to 250 products per search or category. Split a broad search into narrower ones for more.
• Search pages are the hardest to read. The Product Scraper retries blocked pages through a residential exit and a browser. A URL that still fails is reported in the run's status message and OUTPUT record, and failed URLs aren't charged.
• Prices follow the delivery location. Amazon hides the featured price when a product doesn't ship to the request's location. Set a proxy country or a delivery ZIP, and budget for the ZIP: it's charged per product whose page actually showed your location.
• Single-product pages vary. Attributes, bestseller ranks, the description and A+ content aren't always present, and the offers lookup can return an empty list.
• Reading, not buying. The Actors never sign in, so carts, checkout and your Amazon account are untouched.

Spending control

• A charge cap per run. maxTotalChargeUsd bounds the whole run. The Reviews Scraper also stops early when Amazon blocks a dozen product pages in a row, and keeps what it saved.
• Scoped API keys. An unattended agent can get its own key under Settings → API & Integrations → API keys, limited to actors:run and runs:read, with a daily charge limit.
• Prepaid balance. Every run draws from your balance, so a scheduled agent can't create a surprise invoice.

Is it legal to scrape Amazon with an AI agent?

The Actors read public catalog and review pages that any visitor sees without signing in, and Scrape.ly Store Actors collect only publicly available data. Amazon's conditions of use set its own rules on automated access, and how they bind you depends on your location and how you use the data. Product data is business data; reviewers are people, which is why names stay off unless you switch them on. Our guide on how to scrape Amazon covers the details, and this is not legal advice.

For REST access with Python, JavaScript and cURL, see the Amazon API page. For reviews at scale, read our guide to scraping Amazon reviews; for the wider picture, see our guide to ecommerce scraping.

FAQ

What is Amazon MCP?

It's any MCP server that connects an AI agent to Amazon. Amazon runs two of its own: the Amazon Ads MCP Server for advertising partners and the AWS MCP Server for cloud work. For product, price and review data, agents use a scraping server such as Scrape.ly's, which runs Amazon scrapers on public pages.

Does Amazon support MCP?

Yes, for ads and AWS. The Amazon Ads MCP Server has been in open beta since 2 February 2026 for Ads partners with active API credentials, and the AWS MCP Server covers AWS services. Neither is a catalog or review feed.

Is there an MCP for AWS?

Yes. The AWS MCP Server is a managed server in the Agent Toolkit for AWS. Agents search AWS documentation without signing in and use your IAM credentials to call AWS APIs.

Can Claude shop on Amazon?

Claude can research products through the Scrape.ly server: compare prices, ratings and reviews from live pages. It can't buy anything through it, because the Actors never sign in. The shopping server that once ranked for this, Fewsats/amazon-mcp, is discontinued.

How much does an Amazon MCP setup cost?

The Scrape.ly server is free to connect. Products cost $1.06 per 1,000 plus $0.0004 per run, reviews $4.39 per 1,000, platform usage included. Our two keyboard runs came to $0.028 at list price, and the free plan includes $5 every month.

Do I need a seller account or the Product Advertising API?

No. The Actors read public pages, so there are no seller credentials, Ads partner credentials or affiliate API keys to set up. A Scrape.ly account is all you need.