Almost nobody in this industry publishes what their data costs. You request a quote, you get a call, and the number arrives after two meetings shaped by what the vendor thinks you can pay.
Here is ours, with the arithmetic, and a worked example measured against the live API rather than illustrated.
The unit
Magpie's Data API bills per row returned. A row is one SKU in one month, carrying GMV, units sold, price, discount depth, ratings and the full three-level category tree.
Rows bill at 15 credits. Credits cost USD 0.00125 each at the entry tier and get cheaper with larger deposits. So a row is about 1.9 US cents at the worst price you will ever pay.
There is a second rate. A top-N slice — the highest-GMV rows per category, per platform, per month — bills at 50 credits a row. That is 3.3× the unit price and it is almost always the cheaper thing to buy, which sounds contradictory until you see the volumes.
A real export, priced
Shampoo. Shopee. Indonesia. January to June 2026. This is an actual dry-run estimate returned by the API:
374,169 rows · 5,612,535 credits · USD 7,015.67
That is the entire category, every SKU, six months. Now the same scope with "top": 100 — the hundred largest brands by GMV in each month:
600 rows · 30,000 credits · USD 37.50
Same category, same months, same marketplace. USD 7,015 against USD 37.50 — a factor of 187, for a difference in question rather than a difference in data quality.
Which one you actually need
Most questions want the second. "Who leads this category and how has that moved" is a top-N question. So is price architecture, so is entry sizing, so is nearly every brand-level analysis anyone runs.
The full export is for a different job: loading a category into your own warehouse to build a product on top of, or running SKU-level analysis where the long tail is the point. Both are legitimate. Neither is what somebody means when they say they want "the beauty category data".
The failure mode this pricing structure is designed to prevent is buying the first when you needed the second. Which is why:
Every estimate is free
Add "estimate": true to any export request and the API returns the exact row count, the credit cost and the dollar price, and charges nothing. Not an approximation — the same number that will be billed, computed against the same catalogue that delivers the export.
This is not a courtesy. It is the mechanism that makes per-row pricing safe to buy. You can scope, price, adjust and re-price as many times as you like before anything is charged, and the figure you see is the figure you pay.
Failed and empty queries cost nothing either. Credits are charged against rows actually delivered.
What you can buy
Coverage is not uniform and we publish it rather than describing it. Read from the catalogue on 8 September 2026:
Indonesia runs from November 2020 and carries 121 categories across Shopee, Tokopedia, Lazada, Blibli and TikTok Shop. Thailand, Singapore, the Philippines, Vietnam and Malaysia run from November 2022 and carry twelve to fifteen categories each, across Shopee and Lazada — plus TikTok Shop in Thailand, Singapore and the Philippines. Every market's latest month is June 2026.
Indonesia is the deep market by a wide margin, and we would rather say that plainly than let a buyer discover it in diligence.
The scraping side
On-demand collection is priced per request rather than per row. A Shopee product page is 8 credits, a search or merchant listing 6 credits per URL, per-variant sold counts 10 credits per item. Tokopedia and Blibli requests are 1 credit. Everything is charged on success.
The operationally important detail is not the price, it is the gating: product-page fetches run unattended, while search listings, merchant catalogues and variant counts require approval and only process between 08:00 and 18:00 GMT+8. If you are designing a scheduled job, that distinction decides your architecture.
Why publish this
Because the alternative wastes everyone's time. A buyer who can price their own question in five minutes either proceeds or does not, and both outcomes are better than three weeks of scoping calls ending in sticker shock.
The full rate card and coverage table are public, and you can run an estimate against your own scope as soon as you have a key.