Keepa Mcp

keepa-mcp

Amazon price and sales-rank history for your AI assistant, via the Keepa API.

[CI] [npm]

Unofficial package. keepa-mcp is not affiliated with, endorsed by, or supported by Keepa GmbH. "Keepa" and "Amazon" are trademarks of their respective owners. This is an independent open-source client for the publicly documented Keepa API.

What it does

Quick start

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{  "mcpServers": {    "keepa": {      "command": "npx",      "args": ["-y", "keepa-mcp"],      "env": {        "KEEPA_API_KEY": "YOUR_API_KEY_HERE"      }    }  }}

Claude Code

claude mcp add keepa -e KEEPA_API_KEY=your_key -- npx -y keepa-mcp

OpenAI Codex CLI

Add to ~/.codex/config.toml:

[mcp_servers.keepa]command = "npx"args = ["-y", "keepa-mcp"]
[mcp_servers.keepa.env]KEEPA_API_KEY = "YOUR_API_KEY_HERE"

Getting a Keepa API key

Sign up at https://keepa.com/#!api to get your API key.

The free tier provides 100 tokens per minute. Each product data request costs tokens proportional to the data retrieved (history length, number of offers, etc.). Typical product lookups cost 1-5 tokens; best-seller lists and deals cost more. See the Keepa API docs for the full token cost table.

Token economics at a glance:

• get_product: ~2-4 tokens — the product record plus the stats, rating and buybox add-ons it needs. Keepa returns no current price, rating, review count or offer count without them.
• get_price_history / get_sales_rank_history: ~2-4 tokens
• get_best_sellers: 50-100 tokens per request
• get_deals: 5-25 tokens per page
• Free tier refills at 100 tokens/minute

Live offer data (offers) is not requested. It triggers a real-time marketplace lookup costing 10+ extra tokens per call, which on a small bucket can fail the request outright. Versions before 0.1.2 always sent offers=20, so get_product never actually cost the 1 token this table used to claim.

If you hit the quota, the tool returns a clear error message and you can retry after the bucket refills.

Example conversations

1. Product research

"Look up B08N5WRWNW, then show me the last 6 months of price history and tell me whether demand has been growing."

The assistant calls get_product, then get_price_history with range="6m", then get_sales_rank_history with range="6m", and synthesises a verdict.

2. Category opportunity scan

"Pull the top 20 best sellers in Electronics (category 172282), then get current details and sales rank trends for the first three."

The assistant calls get_best_sellers, then get_product and get_sales_rank_history for the first three ASINs, and summarises the demand signals.

3. Deal hunting

"Find highly-rated Amazon deals with at least 25% off, then show me the price history for the top result to confirm it is a genuine dip."

The assistant calls get_deals with min_price_drop_pct=25, min_rating=45, picks the top ASIN, then calls get_price_history with range="6m" to validate the claimed price drop against historical data.

Development

# Install dependenciesnpm install
# Run tests (no live HTTP; all upstream calls are mocked)npm test
# Build TypeScript to dist/npm run build
# Start the server (requires KEEPA_API_KEY in environment)KEEPA_API_KEY=your_key node dist/index.js

Project structure

src/  index.ts                 # Server entry point, tool registration  constants.ts             # API base URL, Keepa time offset, limits  types.ts                 # TypeScript interfaces  services/    keepa-client.ts        # Fetch wrapper, time utilities, history decoder  tools/    get-product.ts    get-price-history.ts    get-sales-rank-history.ts    search-products.ts    get-best-sellers.ts    get-deals.tstests/  keepa-client.test.ts     # Unit tests for time conversion and history decoding  tools.test.ts            # Integration tests with mocked fetch + fixture JSON  fixtures/                # Realistic Keepa response shapes

Built by

Built by Puya Ventures LLC. I build custom MCP servers and AI integrations for product-research, e-commerce, and SaaS teams. Get in touch: [email protected] | Portfolio: puyarahmanian.com

Part of the Product-Research MCP Suite: keepa-mcp · google-trends-mcp · junglescout-mcp

Privacy

This server runs entirely on your machine. It collects no telemetry and stores no data. The only network calls it makes are to the Keepa API (api.keepa.com), sending your API key and the ASINs, keywords, or category ids you ask about. Your API key is read from the KEEPA_API_KEY environment variable and never written to disk or sent anywhere except Keepa. See Keepa's privacy policy: https://keepa.com/#!privacy

License

MIT. See LICENSE.