scavio-amazon

Search Amazon, read full product detail by ASIN, and list every seller offer on an ASIN with the buy-box winner. Clean normalized JSON with price, rating, review count, availability, shipping and sellers. 3 endpoints, all 1 credit, 22 marketplaces.

Amazon via Scavio

Search Amazon by keyword, pull full product detail for an ASIN, and list every seller offering that ASIN including which one holds the buy box. All three endpoints return clean normalized JSON across 22 Amazon marketplaces.

When to trigger

Use this skill when the user asks to:

• Find products on Amazon by keyword, with price, rating and review count
• Look up a specific product by ASIN: price, availability, images, specifications, variants, shipping
• Compare sellers on one product, find the cheapest offer, or see who holds the buy box
• Track a price, check stock, or watch a listing over time
• Research a marketplace other than the US (Germany, UK, Japan, ...)
• Mine best sellers rank, sales volume or badges for competitor research

Three things to read before you call anything

1. There is no sort. At all. Amazon accepts every sort value and ignores all of them. Verified by comparing price ordering across price-asc-rank, price-desc-rank, review-rank and date-desc-rank: identical, unordered result sets every time. There is therefore no sort_by parameter, and sending one gets you a warnings entry, not a sorted list. If the user wants "cheapest first", sort the returned products[] yourself on price — and say that you sorted one page locally, not that Amazon ranked them.

2. country is a two-letter country code, not a domain and not a ZIP. us, gb, de, jp. Two do not match the domain suffix people expect: amazon.com is us (not com) and amazon.co.uk is gb (not uk). The old domain parameter (com, co.uk) still works as a deprecated alias and is translated for you, but write new calls with country. A code that is two letters but not a real marketplace does not fail — it quietly returns the US storefront, so a typo looks like a successful search of the wrong country. Check the code against the list below before sending it.

3. reviews_count from search can be a rounded display value; from product and offers it is always exact. Search carries the count as page text: "(517)" parses to an exact 517, but anything Amazon abbreviates — "(1.3K)", "(92.9K)" — parses to 1300 and 92900, which are Amazon's own rounded figures, not real counts. /product and /offers return the true integer. Never present a large search-derived count as exact, and never diff a search count against a product count and call it a change.

Setup

Get a free API key at https://scavio.dev (50 free credits to get started, no card required):

bash

export SCAVIO_API_KEY=sk_live_your_key

Every request is a POST with a JSON body and:

text

Authorization: Bearer $SCAVIO_API_KEY

Endpoints

Base URL: https://api.scavio.dev. Every data endpoint costs 1 credit.

Endpoint	Credits	Description
POST /api/v1/amazon/search	1	Keyword search: product cards, filters, related searches
POST /api/v1/amazon/product	1	Full detail for one ASIN
POST /api/v1/amazon/offers	1	Every seller offer on one ASIN, with the buy-box winner
GET /api/v1/amazon/options	0	The marketplace list. No API key, no credit.

Workflow

1. Find products: POST /amazon/search with query. Each card already carries asin, price, currency, rating, reviews_count, badge, sales_volume and delivery — for a shortlist you are done in one call.
2. Page through: pass page: 2, page: 3. One page per call, 1 credit per call. There is no way to fetch several pages in one request.
3. Deep-dive: POST /amazon/product with the ASIN for description, features, images, videos, specifications, variants, best sellers rank and structured shipping.
4. Compare sellers: POST /amazon/offers with the same ASIN when the user cares about price, condition or who is selling. has_buy_box and other_sellers_count on the product response tell you whether that call is worth 1 credit.
5. Switch marketplace: set country on any of the three.

/product and /offers take the ASIN in query (the field every Scavio client has always used) or in asin — they are the same parameter, so send one.

Parameters

Search (/api/v1/amazon/search)

Parameter	Type	Default	Description
query	string	required	Search keyword (1-500 chars)
country	string	us	Marketplace, ISO 3166-1 alpha-2 (see below)
page	number	1	1-based results page
domain	string	--	Deprecated alias for country (com, co.uk, de). Translated for you.
start_page	number	--	Deprecated alias for page

No sort_by, no pages, no category_id, no merchant_id, no price filter, no zip_code, no device, no language, no currency. None of those exist upstream any more, so none are accepted. To narrow a search, use the URLs in filters[].options[], or add the qualifier to query ("laptop under 500").

Product (/api/v1/amazon/product)

Parameter	Type	Default	Description
query	string	required	The 10-character ASIN, e.g. B09XS7JWHH. Alias: asin.
country	string	us	Marketplace, ISO 3166-1 alpha-2
domain	string	--	Deprecated alias for country

Offers (/api/v1/amazon/offers)

Parameter	Type	Default	Description
query	string	required	The 10-character ASIN. Alias: asin.
country	string	us	Marketplace, ISO 3166-1 alpha-2
domain	string	--	Deprecated alias for country

Page 1 only. The response echoes page and has_more_pages, but there is no verified upstream parameter to request page 2, so none is exposed. If has_more_pages is true, say the list is the first page rather than implying it is complete.

Marketplaces

us gb de fr it es nl be se pl tr ca mx br jp cn sg in au ae sa eg

GET /api/v1/amazon/options returns the same list with labels, needs no API key and costs nothing. Its body is { domains, countries, languages, currencies }: countries holds {value, label} pairs for the codes above, domains the matching amazon.<suffix> aliases, and languages / currencies are empty arrays — they are kept only so old parsers do not break, since neither is a request parameter any more.

country must be exactly two letters or the request is a 400. A two-letter code that is not on the list above does not error: it is forwarded to Amazon, which quietly serves the US storefront instead — country: "zz" comes back 200 with amazon.com results. The deprecated domain alias does the same, so an old caller sending domain: "uk" (not a real Amazon suffix — the UK is co.uk) also gets the US storefront with no error. Neither case is distinguishable from a real US search once you have the response, so validate the code against /options before sending it rather than trusting what comes back.

Examples

bash

# 1. Keyword search
curl -s -X POST https://api.scavio.dev/api/v1/amazon/search \
  -H "Authorization: Bearer $SCAVIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "wireless headphones"}'

# 2. Page 2 of the German marketplace
curl -s -X POST https://api.scavio.dev/api/v1/amazon/search \
  -H "Authorization: Bearer $SCAVIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "kopfhoerer", "country": "de", "page": 2}'

# 3. Product detail by ASIN
curl -s -X POST https://api.scavio.dev/api/v1/amazon/product \
  -H "Authorization: Bearer $SCAVIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "B0GRVFY42Q"}'

# 4. Every seller offer on that ASIN
curl -s -X POST https://api.scavio.dev/api/v1/amazon/offers \
  -H "Authorization: Bearer $SCAVIO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"asin": "B0GRVFY42Q", "country": "us"}'

# 5. The marketplace list -- free, no key needed
curl -s https://api.scavio.dev/api/v1/amazon/options

Search, then price-check the cheapest seller:

python

import os, requests

BASE = "https://api.scavio.dev"
HEADERS = {"Authorization": f"Bearer {os.environ['SCAVIO_API_KEY']}"}

def call(path, body):
    r = requests.post(f"{BASE}/api/v1/amazon/{path}", headers=HEADERS, json=body)
    r.raise_for_status()
    return r.json()

# 1. Search. Results are in Amazon's default ranking -- there is no sort param.
#    Sort locally if the user wants cheapest first, and say so.
res = call("search", {"query": "wireless headphones", "country": "us"})["data"]

for p in sorted(res["products"], key=lambda p: p["price"] or float("inf")):
    # reviews_count here is rounded when Amazon abbreviated it: "(1.3K)" -> 1300
    print(p["asin"], p["price"], p["currency"], p["rating"], p["reviews_count"], p["title"][:60])

asin = res["products"][0]["asin"]

# 2. Detail. price/list_price are numbers, currency is a separate string.
d = call("product", {"query": asin})["data"]
print(d["title"], d["price"], d["currency"], d["availability"], d["sold_by"])

# 3. Only pay for offers when there are other sellers to compare.
if (d["other_sellers_count"] or 0) > 0:
    offers = call("offers", {"asin": asin})["data"]
    for o in sorted(offers["offers"], key=lambda o: o["price"] or float("inf")):
        tag = "buy box" if o["is_buy_box_winner"] else ""
        print(o["seller_name"], o["condition"], o["price"], o["shipping_price"], tag)
    if offers["has_more_pages"]:
        print("first page of offers only")

Response shapes

Every response uses the envelope { data, response_time, credits_used, credits_remaining }. A warnings array is added at the top level only when the request carried a retired parameter — see Failure handling.

Fields mean the same thing on all three endpoints: title, price (a number), currency (a separate string), rating (0-5), reviews_count, image. Any field with no upstream equivalent is absent, never a permanent null.

Search

data is { query, page, total_results, total_results_text, count, products[], filters[], related_searches[] }.

json

{
  "data": {
    "query": "laptop",
    "page": 1,
    "total_results": 100000,
    "total_results_text": "1-16 of over 100,000 results for laptop",
    "count": 16,
    "products": [
      {
        "asin": "B0GRVFY42Q",
        "title": "HP 15.6\" FHD Laptop 2026 Edition, Intel Processor, 8GB RAM, 256GB SSD",
        "url": "https://www.amazon.com/HP-Laptop-Intel-Processor-256GB/dp/B0GRVFY42Q/ref=sr_1_1?dib=...",
        "image": "https://m.media-amazon.com/images/I/71sTXs4lkqL._AC_UY218_.jpg",
        "price": 414.99,
        "currency": "USD",
        "rating": 4.2,
        "reviews_count": 517,
        "is_sponsored": false,
        "position": 3,
        "badge": "Overall Pick",
        "sales_volume": "1K+ bought in past month",
        "delivery": { "is_free": true, "date": "Tomorrow, Aug 1", "fastest_date": "Aug 3 - 4" }
      }
    ],
    "filters": [
      {
        "name": "Popular Shopping Ideas",
        "options": [
          {
            "name": "Thinkpad",
            "url": "https://www.amazon.com/s?k=laptop+thinkpad&ref=sr_nr_p_rag_integrated_qb_0",
            "refinement": null
          }
        ]
      }
    ],
    "related_searches": [
      { "position": 1, "query": "macbook", "url": "https://www.amazon.com/s?k=macbook&ref=rsl_sug_0_0" }
    ]
  },
  "response_time": 4940,
  "credits_used": 1,
  "credits_remaining": 999
}

Read these carefully before building on them:

• total_results is a floor parsed out of the page text: "over 100,000 results" becomes 100000. total_results_text is the sentence it came from. Never divide either by a page size to compute a page count — page until products[] comes back short.
• count is how many products this page returned, typically 16 and not guaranteed.
• position is Amazon's own grid index including ad and carousel slots, so it is neither the array index nor contiguous — a live page returned 3, 4, 5, 8, 9, 10, ..., 22. It is passed through untouched. Use the array order for "the Nth result".
• price and rating can be null on a card Amazon rendered without them. Guard before arithmetic.
• filters[].options[].refinement is Amazon's refinement token. There is no parameter to send it back to, so use the option url instead.
• is_sponsored marks paid placements. Exclude them when the user asks for organic results.

Product

data is { asin, title, brand, url, description, features[], price, list_price, currency, rating, reviews_count, is_prime, is_sponsored, has_buy_box, availability, max_quantity, sold_by, other_sellers_count, sales_volume, climate_pledge_friendly, image, images[], videos[], best_sellers_rank[], categories[], specifications{}, variants[], shipping{}, reviews[] }.

json

{
  "data": {
    "asin": "B0GRVFY42Q",
    "title": "HP 15.6\" FHD Laptop 2026 Edition, Intel Processor, 8GB RAM, 256GB SSD",
    "brand": "HP",
    "url": "https://www.amazon.com/dp/B0GRVFY42Q",
    "description": "Your Reliable Companion for Learning, Work & Everyday Computing...",
    "features": ["Powerful Everyday Performance: Intel N100 processor with 8GB RAM..."],
    "price": 414.99,
    "list_price": 799,
    "currency": "USD",
    "rating": 4.2,
    "reviews_count": 517,
    "is_prime": true,
    "is_sponsored": false,
    "has_buy_box": true,
    "availability": "In Stock",
    "max_quantity": 30,
    "sold_by": "Omnitech Global",
    "other_sellers_count": 0,
    "sales_volume": "1K+ bought in past month",
    "climate_pledge_friendly": true,
    "image": "https://m.media-amazon.com/images/I/71sTXs4lkqL._AC_SY300_SX300_QL70_FMwebp_.jpg",
    "images": ["https://m.media-amazon.com/images/I/71sTXs4lkqL._AC_SL1500_.jpg"],
    "videos": [
      {
        "title": "HP 15.6\\\" FHD Laptop 2026 Edition with Copilot AI",
        "url": "https://m.media-amazon.com/images/S/vse-vms-transcoding-artifact-us-east-1-prod/de91.../default.jobtemplate.hls.m3u8",
        "thumbnail": "https://m.media-amazon.com/images/I/51YqUDn8TkL.SX522_.jpg",
        "duration_seconds": 12,
        "width": 1920,
        "height": 1080
      }
    ],
    "best_sellers_rank": [{ "category": "Computers & Accessories", "rank": 98 }],
    "categories": [
      {
        "name": "Electronics",
        "url": "https://www.amazon.com/electronics-store/b/ref=dp_bc_1?ie=UTF8&node=172282",
        "node": "172282"
      }
    ],
    "specifications": { "CPU Model Number": "Intel Processor N100", "Screen Size": "15.6 inches" },
    "variants": [
      { "asin": "B0GRVFY42Q", "dimensions": { "Color": "Natural Silver" }, "is_selected": true, "is_available": true }
    ],
    "shipping": {
      "is_prime": true,
      "zipcode": "10001",
      "options": [
        {
          "type": "free",
          "price": 0,
          "is_prime": true,
          "delivery_date": "Tomorrow, August 1",
          "order_deadline_hours": 4,
          "order_deadline_minutes": 55
        }
      ]
    },
    "reviews": [
      {
        "id": "R25NUR3YMMFTG1",
        "author": "Domika0717",
        "date": "Reviewed in the United States on June 7, 2026",
        "verified_purchase": true
      }
    ]
  },
  "response_time": 2768,
  "credits_used": 1,
  "credits_remaining": 998
}

• availability is free text in the marketplace's language ("In Stock", "Nur noch 3 auf Lager"). There is no boolean in-stock flag, because the wording cannot be parsed reliably across 22 storefronts. Quote it; do not derive from it.
• reviews[] is metadata only — id, author, date, verified flag. Upstream carries no review body and no per-review rating anywhere on the product page. Never claim to have read a review's text.
• specifications and variants[].dimensions use Amazon's own human labels as keys ("CPU Model Number", "Color"), kept verbatim. They differ by category and by marketplace, so look keys up defensively.
• video.title arrives double-escaped from upstream (HP 15.6\\" FHD ...). It is passed through as sent rather than "fixed", because un-escaping would corrupt legitimate backslashes.
• image is a resized thumbnail; images[] holds the full-size set.
• other_sellers_count is the cheapest way to decide whether an /offers call is worth a credit.

Offers

data is { asin, title, image, rating, reviews_count, note, count, total_offers, has_more_pages, page, offers[] }.

json

{
  "data": {
    "asin": "B0GRVFY42Q",
    "title": "HP 15.6\" FHD Laptop 2026 Edition, Intel Processor, 8GB RAM, 256GB SSD",
    "image": "https://m.media-amazon.com/images/I/412kavhXHzL.jpg",
    "rating": 4.2,
    "reviews_count": 517,
    "note": "Currently, there are no other sellers matching your location and / or item specification.",
    "count": 1,
    "total_offers": 1,
    "has_more_pages": false,
    "page": 1,
    "offers": [
      {
        "condition": "New",
        "seller_id": "A171014CP909KT",
        "seller_name": "Omnitech Global",
        "ships_from": "Omnitech Global",
        "is_fulfilled_by_amazon": false,
        "is_buy_box_winner": true,
        "is_prime": true,
        "is_national_prime": true,
        "price": 414.99,
        "currency": "USD",
        "list_price": 799,
        "shipping_price": 0,
        "discount_percentage": 48.06,
        "discount_amount": 384.01,
        "price_notice": null,
        "promotion": null,
        "quantity": 30,
        "delivery": { "min_hours": 24, "max_hours": 72, "date": "August 3 - 4", "is_free": true },
        "prime_delivery": { "date": "Tomorrow, August 1", "order_deadline": "Order within 4 hrs 55 mins" }
      }
    ]
  },
  "response_time": 3830,
  "credits_used": 1,
  "credits_remaining": 997
}

• price is the item price and shipping_price is separate. The landed cost is price + shipping_price; compute it, do not assume price includes delivery.
• is_buy_box_winner marks the offer Amazon serves by default. It is not always the cheapest — say which one is which.
• count is the offers on this page; total_offers is what upstream reported. They agreed in every observed response, but trust count for anything you are about to iterate.
• note is genuine user-facing text explaining an empty or short list. Surface it instead of reporting "no offers found".
• price_notice and promotion were null in every offer observed so far. Treat a non-null value as a bonus, never a required field.
• condition is Amazon's own wording ("New", "Used - Very Good"). Do not normalize it into a grade you invented.

Guardrails

• Every call to /search, /product and /offers costs 1 credit, including a search that returns nothing. Tell the user before paginating deeply or looping /product over a whole results page. /options is free.
• Never fabricate a product title, ASIN, price, rating, review count, seller or availability. Only return API data. If a field is null, say it is unavailable.
• Never claim results are sorted by Amazon. There is no sort parameter. If you ordered them, say you sorted one page of results locally.
• Prices are numbers in currency, which changes with country. Never mix currencies in one comparison and never convert between them — you have no exchange rate.
• Search reviews_count is rounded whenever Amazon abbreviated it (1.3K -> 1300); product and offers counts are always exact. Do not mix the two in one comparison.
• position is Amazon's grid index including ads. Do not present it as "rank 3 of the organic results".
• Always include the product url so the user can verify. For a search card that is the ref-tracked link Amazon serves; it is the real result URL and works as-is.
• Prices and delivery estimates are point-in-time and vary by marketplace and location. Say when the data was fetched rather than implying it is stable.
• /offers returns page 1 only. If has_more_pages is true, do not present the list as every seller.
• For a price comparison across two products, run two searches or two product calls — 2 credits — and present both. Never infer the second from the first.

Failure handling

• 400 — an invalid parameter: an empty or over-500-character query, a country that is not exactly two letters, or a /product or /offers call with no ASIN in either query or asin. Not billed. Fix the request.

• 401 — the API key is invalid or missing. Check SCAVIO_API_KEY.

• 429 — rate or usage limit exceeded. Wait before retrying. See https://scavio.dev/docs/rate-limits.

• 502 — Amazon data is temporarily unavailable, or the ASIN could not be fetched from that marketplace. Not billed. Wait a few seconds and retry once; if it repeats, check the ASIN exists on that country's storefront before retrying again.

• 503 — the upstream fetch never completed (network failure or timeout). Not billed. Retry after a short backoff.

• A 200 with a top-level warnings array means the request carried a parameter that no longer exists and was ignored. Read it and fix the caller, because the response is not filtered the way the request implied:

  json

  {
    "data": { "query": "laptop", "page": 1, "count": 16 },
    "response_time": 4512,
    "credits_used": 1,
    "credits_remaining": 996,
    "warnings": [
      "sort_by is no longer supported: the upstream marketplace ignores every sort value and returns the default ranking. Results are unsorted."
    ]
  }
  

  Warned parameters: all nine retired ones — sort_by, pages, category_id, merchant_id, language, currency, device, zip_code, autoselect_variant. The data object is unchanged and the call is still billed.

• An empty products[] on a 200 means Amazon returned no matches for that keyword in that marketplace. Try a broader keyword or a different country; do not retry the same request.

• Results from the wrong storefront mean the marketplace code did not resolve. Both an off-list two-letter country and an unrecognised legacy domain fall back to us silently rather than erroring, so a typo returns plausible US data instead of an error. Check the code against /api/v1/amazon/options and prefer country over domain.

• Calls typically take 3-5 seconds. Set a client timeout of at least 60 seconds.

• If SCAVIO_API_KEY is not set, prompt the user to export it before continuing.

Related

• scavio-walmart — the same product-research shape for Walmart.
• scavio-google-shopping — cross-retailer price comparison when the user is not tied to Amazon.
• Full API reference: https://scavio.dev/docs/amazon-api (one page per endpoint: amazon-api, amazon-product, amazon-offers)

LangChain

bash

pip install langchain-scavio

python

from langchain_scavio import ScavioAmazonOffers, ScavioAmazonProduct, ScavioAmazonSearch

tools = [ScavioAmazonSearch(), ScavioAmazonProduct(), ScavioAmazonOffers()]

All three endpoints have a LangChain tool; each reads SCAVIO_API_KEY from the environment.