Amazon.com, from the command line

Pulling structured data out of Amazon usually means a pile of brittle scrapers, one per page type, each breaking the next time a selector moves. amz puts all of it behind one tool with sensible defaults, real output formats, and pipelines that compose.

amz product B075F5X8BR                 # one product, fully normalized
amz search "mechanical keyboard" -o jsonl
amz reviews B075F5X8BR --stars 1 -o csv
amz bestsellers electronics            # the live top-100 chart

It reads the public pages on amazon.com over plain HTTPS, so there is nothing to sign up for to get started. The binary is pure Go with no runtime dependencies and nothing to install alongside it: the local store is SQLite through a pure Go implementation, so there is no external binary that can be missing.

robots.txt is fetched from the marketplace and enforced before every request, requests are paced at 3s with a floor no flag can lower, and no page is ever requested behind a login. Where that means a command returns less than you might hope, it says so on the record rather than filling the gap in.

What you can do with it

• Look up products. amz product fetches a detail page and normalizes the title, the brand and its storefront, the price and what it was struck from, the rating and its histogram, availability, the buy box and who wins it, feature bullets, technical detail, images and their variants, the breadcrumb, the variation matrix, and every Best Sellers Rank into one record. Each field records which of Amazon's own named regions it came out of.
• Search the catalog. Stream result cards with the refinement vocabulary read off the page itself, real pagination that stops where Amazon says the results do, and --all to partition a query and get past the 306 result ceiling.
• Read the social proof. The reviews and the rating histogram the detail page carries, and the answered question count with any inline pairs. The full corpus is behind a login and amz says so instead of pretending otherwise.
• Compare offers. The buy box, who sells it, who ships it, and how many offers sit behind the panel Amazon builds with javascript.
• Walk the charts and trees. Bestsellers, new releases, movers and shakers, most wished for, and most gifted, plus category browse nodes, brand storefronts, seller profiles, author pages, and today's deals.
• Keep what you read. A resumable crawl into a local SQLite store, a graph of sixteen predicates over what it recorded, SQL and full text search with no network, and an export to JSONL or RDF.
• Hand it to a model. amz serve puts the read commands behind HTTP and amz mcp puts the same registry on stdio as Model Context Protocol.

Where to go next

• New here? Start with the introduction for the mental model, then the quick start.
• Want to install it? See installation.
• Looking for a specific task? The guides cover products, search, reviews and Q&A, offers, charts, crawling at scale, and a page of recipes.
• Need the exact shape of a record? The data model names every field of every surface.
• Need every flag? The CLI reference is the full surface.
• Getting less than you expected? Troubleshooting and amz why explain each case, with the measurement and the date behind it.