Skip to content
amz

Recipes

End-to-end pipelines that combine amz commands into real work: price tracking, review mining, competitive scans, and market research.

amz is built to compose. Every command streams the same record types through the same renderer, so the output of one is the input of the next. These recipes chain them into the work people actually do with Amazon data. They use jq for JSON wrangling, but plain sed/awk versions are shown where they are simpler.

Track a price over time

Append a timestamped price row to a CSV on every run, then watch the file:

amz price B075F5X8BR -o jsonl \
  | jq -r '[now|todate, .asin, .price.value, .price.currency] | @csv' \
  >> price_log.csv

Drop that line in a cron job and you have a price history with no database. To watch a basket, loop a file of ASINs:

while read asin; do amz price "$asin" -o jsonl; done < watchlist.txt \
  | jq -r '[now|todate, .asin, .price.value] | @csv' >> basket_log.csv

Read the buy box and how many offers sit behind it

The other sellers' rows are drawn by JavaScript and the endpoint behind them answers 404 to a direct request, so amz offers returns the buy box winner and the count rather than a list. Take both together:

amz offers B075F5X8BR -o jsonl \
  | jq -c '{price, currency, condition, seller_name, fulfilled_by, is_buybox}'

{"price":209378,"currency":"VND","condition":"","seller_name":"Amazon.com","fulfilled_by":"Amazon.com","is_buybox":true}

The count is on the product record, beside a complete flag that is never omitted so a partial answer cannot be mistaken for a whole one:

amz product B075F5X8BR -o json | jq -c '.[0].other_offers'

{"loaded":1,"total_count":2,"complete":false,"url":"https://www.amazon.com/gp/offer-listing/B075F5X8BR/ref=dp_olp_NEW_mbc?ie=UTF8&condition=NEW"}

Two offers exist, amz holds one, and it says so. amz why offers has the measurement and the date. --condition and --prime still filter what comes back, which matters when the buy box is a used copy and you only want new:

amz offers B075F5X8BR --condition new --prime -o jsonl

Enrich a chart into full product records

Charts give you ASINs and a thumbnail of data. Fan each one out into a full product record:

amz bestsellers electronics -n 25 --fields asin -o csv --no-header \
  | xargs -I{} amz product {} -o jsonl > top25.jsonl

Now ask questions of the file. Average discount among the top 25:

jq -s 'map(.offer.savings_pct // 0) | add / length' top25.jsonl

The brands that appear most:

jq -r '.brand.name // ""' top25.jsonl | sort | uniq -c | sort -rn

Mine the reviews of a product

The corpus is behind a sign-in and what is public is the histogram plus a medley of about a dozen reviews. For the whole rating distribution, read the histogram:

amz product B075F5X8BR -o json | jq '.[0].distribution'

For the medley, sorted by how many people found it helpful:

amz reviews B075F5X8BR -o jsonl \
  | jq -rs 'sort_by(-.helpful_votes) | .[] | "\(.helpful_votes)\t\(.title)"'

Reviews that mention a keyword:

amz reviews B075F5X8BR -o jsonl | jq -r 'select(.text | test("battery"; "i")) | .title'

Compare two products side by side

for a in B075F5X8BR B09B8V1LZ3; do amz product "$a" -o jsonl; done \
  | jq -r '[.asin, .offer.price.value, .rating, .ratings_count, (.ranks[] | select(.overall) | .rank)] | @tsv' \
  | column -t

Scan a search for the best-rated value

Search with refinements, then pick the highest-rated card under a price:

amz search "mechanical keyboard" --stars 4 -n 100 -o jsonl \
  | jq -s 'map(select(.price.value < 120)) | sort_by(-.rating) | .[0:5]'

Get more than 306 results for a query

A plain search tops out at 306 results however many Amazon claims. --all partitions the query on a refinement group and unions the cells on ASIN. Price it first, because it is one search per cell:

amz search "usb-c hub" --all --dry-run
amz search "usb-c hub" --all -o jsonl > hubs.jsonl

Measured on 2026-08-18 that partitioned into 68 cells and returned 1,508 unique results. The stderr summary names the cells that still hit the ceiling and the ones Amazon served unfiltered, so you know where the union is short:

amz search "usb-c hub" --all -o jsonl 2> hubs.log > hubs.jsonl

Walk a brand's catalog

Turn a brand's featured ASINs into full records:

amz brand anker --featured --fields asin -o csv --no-header \
  | xargs -I{} amz product {} -o jsonl > anker.jsonl

Build a dataset with the local store

For anything beyond a one-shot, let the frontier and the local store carry the work. Seed a category's bestsellers, drain the queue, then query with SQL. See crawling at scale for the full treatment.

amz crawl --chart bestsellers --category electronics --limit 100 --depth full
amz query "select json_extract(json, '$.brand.name') brand,
                  count(*) n,
                  round(avg(json_extract(json, '$.offer.price.value'))) avg_price
           from product group by brand order by n desc limit 20"

Cross-marketplace price gap

The same ASIN, priced in two storefronts:

for m in us uk de; do
  amz price B075F5X8BR -m "$m" -o jsonl
done | jq -r '[.marketplace, .price.value, .price.currency] | @tsv'

Dry-run before a big crawl

See exactly which URLs a run would hit, across marketplaces, without fetching:

amz product B075F5X8BR -m jp --dry-run
amz bestsellers electronics -m de --dry-run

Keep iterating for free

Every successful fetch is cached, so once you have pulled a page you can refine the shape of the output as much as you like without touching the network:

amz product B075F5X8BR -o json                       # first run hits the network
amz product B075F5X8BR --fields asin,price,ranks     # served from cache
amz product B075F5X8BR --template '{{.title}} is #{{.rank}}'
5 min read