Short answer
A language model alone guesses at prices and stock, and its knowledge is months old. Give it tools that read Amazon’s live pages and it can research like a careful shopper: search, shortlist, read reviews, check who is selling, and explain its pick. The open-source amazon-data-api-mcp server adds 12 such tools to Claude and other MCP clients.
Set it up
- Get a free key for the Amazon Data API on RapidAPI. The free plan has 100 requests a month.
- Add the MCP server to your AI app. It runs locally with
npx.
claude mcp add amazon-data -e RAPIDAPI_KEY=your-rapidapi-key -- npx -y amazon-data-api-mcp{
"mcpServers": {
"amazon-data": {
"command": "npx",
"args": ["-y", "amazon-data-api-mcp"],
"env": { "RAPIDAPI_KEY": "your-rapidapi-key" }
}
}
}The MCP server page lists all 12 tools and more clients.
One research task, step by step
Find USB-C chargers of 65 W or more on amazon.de under 40 EUR with at least 4.5 stars.
Shortlist the top 3. For each one:
- read the review summary and list the two most common complaints
- check every seller offer and flag any that are not sold or shipped by Amazon
Finish with a table and say which one you would pick for travel, and why.What the agent calls to answer it:
| Tool | Calls | Why |
|---|---|---|
search_products | 1 | Keyword search on amazon.de, sorted and filtered by price |
get_product | 3 | Wattage, rating, Best Sellers Rank and specs for the shortlist |
get_reviews | 3 | Star breakdown, “Customers say” summary and aspect sentiment |
get_offers | 3 | Every seller’s price, condition, FBA and Prime status |
That’s 10 requests. The reviews step is where an agent beats a search box. It reads the aspect tags, such as “heat” or “cable quality” with mention counts, and the star breakdown, instead of trusting the average rating.
More agents to build
| Agent | How it works |
|---|---|
| Gift finder | Ask for gifts under $50 for a cyclist. The agent reads Gift Ideas and Most Wished For lists with get_best_sellers. |
| Deal watcher | Ask which of today’s deals in Electronics are more than 30% off and over 4 stars, using get_deals. |
| Seller vetting | Before you buy from a third-party seller, the agent pulls their rating, feedback counts and business address with get_seller_profile. |
| Barcode check | Paste a UPC from a store shelf. The agent finds the Amazon listing with lookup_product and compares the price. |
| Cross-border price check | Ask for the same ASIN on amazon.com, amazon.co.uk and amazon.de. Prices come back in each marketplace’s currency. |
Prompting tips
- Name the marketplace. “On amazon.co.uk” sets
country=GB, and prices come back in GBP. - Ask for a shortlist before details. Searching wide, then reading 3 products in depth, uses fewer requests than reading 10.
- Ask it to cite. “Link each product page and say which tool each number came from” keeps the agent from filling gaps with guesses.
- Affiliate sites: check your Amazon Associates agreement before you publish data from a source other than Amazon’s own API (what changed with PA-API).
Amazon Data API
12 endpoints, 23 marketplaces, free plan of 100 requests a month.