Alibaba Scraper

Search Alibaba and get back what a sourcing decision actually needs: the wholesale price band, the minimum order quantity, how many units have sold, the supplier's scores, and how long that supplier has been paying for Gold Supplier status.

48 products per search page, in US dollars, as structured rows.

What each product gives you

• title, url and image_url
• price_text exactly as printed, plus price_min and price_max as numbers. Alibaba quotes a band, not a price: $0.42-0.65 means 42 cents at the largest order quantity and 65 cents at the smallest. Both ends are kept because the spread is the negotiating room.
• moq and moq_text — the minimum order, as a number and as written
• units_sold, review_score, review_count, product_score, shipping_score and supplier_service_score
• company_name, company_country, company_url and company_id
• gold_supplier_years — years of paid Gold Supplier status, which is the cheapest available proxy for how long a supplier has been serious
• is_ad, so paid placements can be filtered out of any ranking you build
• certifications, where the card shows them
• exit_country, which matters more here than on most sites (see below)

Prices are forced to US dollars, on purpose

Alibaba prices in the currency of whoever is asking. During development the same query came back in Singapore dollars, pounds sterling, Omani rials and Pakistani rupees depending on where the request came from, with no error and no indication anywhere in the page. A scraper that ignores this produces a price column that is quietly a mix of four currencies.

Two things stop that here. Every search asks for USD explicitly, and every row records both the printed currency and a normalised currency_code. The run summary counts currencies and raises a warning only when the codes genuinely differ, so US$ and $ do not trigger a false alarm on clean data.

If you see the warning, the prices in that run are not comparable and should not be averaged.

How it handles Alibaba's throttling

This is the part worth understanding before you plan a large run.

Alibaba stops serving a given network after roughly one request. Measured during development: five separate countries each returned a complete page on a first request, and all five were serving a small placeholder thirty seconds later. Retrying the same route does not recover it — sixty-eight consecutive retries through one country returned the placeholder sixty-eight times, while a single request through an untouched one worked immediately.

So this Actor does the opposite of retrying. Each page is fetched through a different country, a country that returns the placeholder is marked spent for the rest of the run rather than tried again, and the run summary reports which countries were used and which were consumed.

The practical consequence is a volume ceiling, not a reliability problem: a handful of pages per run works well, a hundred does not. If a run reports pages it could not read, wait a few minutes and run it again. Recovery is on the order of minutes.

Input

• Search terms — usb cable, led strip, or a full Alibaba search URL pasted as-is
• Result pages per term — each page is 48 products and consumes one country
• Exits to try per page — how many countries to try before giving up on a page. Raise it if pages are failing, lower it to finish faster.
• Preferred country — tried first, then the run falls through to others
• Maximum products per term and maximum search terms

Run summary

Products returned, how many carried a price, how many quoted a band rather than a single figure, how many carried a minimum order, how many were sponsored, the number of distinct suppliers, the lowest and highest prices seen, the currency breakdown, and the exit countries used and spent.

That last pair is the health check for a run. Many countries spent for few products means the pool was already warm, and the fix is to wait rather than to change anything.

What people use this for

Sourcing and cost discovery. The price band plus the minimum order is the whole basis of a landed-cost estimate, and both come through as numbers rather than as text to re-parse.

Supplier shortlisting. gold_supplier_years, units_sold, review_count and the three score fields together separate an established manufacturer from a listing that went up last month. Filtering is_ad out first is usually the right first move, since paid placements sit at the top regardless of merit.

Competitor cost benchmarking. If a competitor sells a product you can identify on Alibaba, the band tells you roughly what they pay for it.

Catalogue seeding. product_id, title, image_url and url are enough to start a catalogue, and the id is stable across runs.

Notes

Prices, stock and supplier scores are read at the moment of the fetch and move constantly, so two runs hours apart will legitimately disagree.

Search highlighting is stripped from titles: Alibaba wraps matched words in markup, and a title with tags in it breaks every downstream use of the field.

No login, no cookies and no API key are needed.