Wikipedia & Wikidata API — Articles, Search & Pageviews
Pricing
from $0.36 / 1,000 articles
Wikipedia search, article text (intro or full), descriptions, categories and Wikidata ids; daily human pageviews and a day's top 1,000; Wikidata entities with flattened properties. Any language edition, no API key, no browser. Diagnostics and empty runs are free.
The Wikipedia edition to read, as its subdomain: en, de, fr, es, ja, zh, pt, simple … Every edition works. Used for search, titles, categories, pageviews, the top list and the language of Wikidata labels. A title given as a URL keeps the language of its own URL.
Full-text Wikipedia searches, one per entry, e.g. ["web scraping"]. Each hit is one search-result row with its snippet, size, word count and last-edit time, up to maxResults per query.
[]Articles to fetch, as titles (Web scraping, Web_scraping) or URLs (https://de.wikipedia.org/wiki/Screen_Scraping). Each becomes one article row. Duplicates and redirects to the same page are fetched and charged once.
[]Categories whose articles to fetch, e.g. Category:Web scraping (the prefix is optional). Only the category's own articles are read — subcategories are not descended into — up to maxResults per category, and each becomes an article row.
[]The most rows one query, category, top-views day or Wikidata search may return. Searches page 50 at a time, categories 500 at a time.
intro returns the lead section as plain text (titles are fetched twenty per request). full returns the whole article as plain text, up to maxTextChars (one request per article). none returns no article rows at all — use it when you only want search results or pageviews.
Also return an article row (with the text chosen above) for every search hit. Adds one request per 20 hits for intros, one per hit for full text.
Cap on text. A longer article is cut here and marked textTruncated: true; textChars and wordCount still describe the whole extract.
The article's visible categories (up to 50; hidden maintenance categories are excluded). No extra request.
The table of contents as sections[] { level, title, anchor, index, number }. One extra request per article.
The first 500 article links, first 100 external links and the number of pages linking here (first 100). linksCountIsPartial is true when any list had more. One extra request per article.
The last 20 revisions as lastRevisions[] { revid, timestamp, comment, size }. Editor usernames are never read or returned, and names inside edit summaries are replaced with [editor]. One extra request per article.
Daily pageviews for every article in the run (titles, category members, and search hits when articles are fetched for them): one pageviews row per title with the series, total, daily average and peak day. Human traffic only (agent user): crawlers and bots are excluded. last30 and last90 end yesterday, the newest day the API has.
First day of the range window, YYYYMMDD or YYYY-MM-DD. Data starts on 2015-07-01.
Dates (YYYY-MM-DD) whose most-read articles to return: up to 1,000 per day from the pageviews API, one pageviews row each with rank and views, capped by maxResults. The main page, Special: pages and other non-article pages are left out unless you tick the box below.
[]Keep Main_Page, Special:Search, Wikipedia:, File:, Portal: and similar entries in a top list.
Entities to fetch, e.g. ["Q42", "Q95"] or https://www.wikidata.org/wiki/Q42. Each becomes one entity row with its label, description, aliases, Wikipedia link and the properties below.
[]Names to look up on Wikidata, e.g. ["Apify"]. Up to maxResults matching entities per name, each an entity row.
[]Wikidata properties to flatten into properties, as ids (P31) or the bundled names (instanceOf). Item values come back as { id, label }, dates at their real precision, images and logos as Commons URLs. Contact details (email, phone, address, personal social handles) are never extracted.
[ "P31", "P17", "P131", "P159", "P571", "P856", "P1128", "P452", "P112", "P169", "P1454", "P625", "P18"]Also return every statement in Wikidata's own JSON as claimsRaw (large: Q42 has 354). Contact-detail properties are removed from it too.
Requests in flight at once, across everything the run is doing. Requests also start at least 100 ms apart, so a run never exceeds 10 requests a second.
Stop after this many seconds. Rows already returned are kept and a free deadline row says what was left.