The world's encyclopedia as one clean table

Wikipedia is 60M+ articles across 300+ languages — the most cited source on the internet, and a pain to consume programmatically if you don't want to write a MediaWiki client. The public REST + Action APIs are keyless, but paginate, localize, and return nested JSON that changes shape per language.

The Wikipedia Article Scraper flattens that into typed rows in three modes: search, single-article extract, and random sampling.

{ "mode": "search", "query": "artificial intelligence", "searchLimit": 10, "language": "en" }

Search returns results with their plain-text extract cleaned of markup:

{
  "title": "Artificial intelligence",
  "description": "Intelligence of machines",
  "extract": "Artificial intelligence (AI) is the capability of computational
    systems to perform tasks typically associated with human intelligence…"
}

Why this over the website or raw API

No API key, no parsing the MediaWiki response variants by hand, no getting blocked for hammering the API. You pass a query (or a title, or "random"), and get back ready-to-load rows. It honours Wikipedia's public API rate expectations and is built for pull-sizes that won't trip them.

Use it for

RAG / LLM grounding corpora, reference lookups, entity descriptions, dataset sampling in any of 300+ languages. The extract is already cleaned — you're feeding text, not HTML, into your pipeline.