There are two ways to get Shopee data, and choosing the wrong one costs either weeks of engineering or a great deal of money.
This is the honest comparison, including where building it yourself is genuinely the right answer.
Option one: build it
The collection is the easy part, and that is what misleads people. A prototype that pulls a product page works in an afternoon.
What follows is the expensive part, and it never ends: pages change shape, and your parser breaks silently rather than loudly. Fields you depend on get renamed. Brand names arrive inconsistently — the same brand spelled four ways across listings — so anything aggregated by brand needs a labelling layer you now maintain. The same product appears under multiple listings and item ids, so totals double-count unless you deduplicate. Categories move, and a taxonomy you built in March no longer maps in September.
None of this is hard. All of it is permanent, and it lands on whoever built it, forever.
Build it if the data is your product, you need something nobody sells, or your volumes are large enough that per-request pricing stops making sense. Those are real cases.
Option two: call an API
You get structured JSON from an endpoint. Someone else absorbs the maintenance.
With Magpie there are two, and the distinction matters more than the price:
On-demand collection for current state. A product page fetch is 8 credits — about one US cent — and returns price, discount, stock and ratings. Search and merchant listings are 6 credits per URL. Per-variant sold counts are 10 credits per item, and only charge on approval.
A maintained monthly panel for history. Already cleaned, brand-labelled and deduplicated, billed per row returned. Indonesia goes back to November 2020 with 121 categories; the other five Southeast Asian markets start November 2022.
Most people who think they want scraping actually want the panel. "Who leads this category and how has that moved" is a history question, and answering it by scraping means collecting for six months before you have anything.
The gating detail nobody mentions until you hit it
If you plan to run collection on a schedule, this decides your architecture.
Product-page fetches run unattended, at any hour. Search listings, merchant catalogues and variant sold counts are approval-gated and process only between 08:00 and 18:00 GMT+8.
You cannot hang a 3am job on the second group. Discovering that after building the scheduler is a bad afternoon, and it is the single most useful thing to know before you start.
What it costs, concretely
Monitoring 400 competitor SKUs nightly is 3,200 credits a day — roughly USD 120 a month.
On the panel side, a measured example: six months of one Indonesian category came to 374,169 rows and USD 7,015.67 in full, against 600 rows and USD 37.50 for the same scope at the top 100 brands per month. A factor of 187, for a difference in question rather than data quality — which is why every export accepts a free dry-run estimate that returns the exact figure before anything is charged.
The comparison worth making
Against general-purpose scraping platforms, the difference is not price, it is what arrives. They will fetch any site you point them at, which is genuinely more flexible. What comes back is raw and yours to clean.
If the data you need is Southeast Asian marketplaces specifically, most of the work you would be paying an engineer to do — the labelling, the deduplication, the taxonomy — is already done here.
If you need arbitrary sites worldwide, use them. That is a real distinction and it points both ways.
Coverage, the full rate card and ten ready-to-run recipes are on the Data API page. Estimates are free, so you can price your exact question before deciding anything.