The difference from a generic page fetch is that you choose the size of what comes back — a query-shaped summary, one section, a plain-text extract or the raw wikitext — so a long article does not have to arrive whole. Parallel search runs several queries at once, which turns a comparison question into one round trip instead of five.
A TypeScript server deployed on Vercel that puts Wikipedia behind MCP: ranked search, full articles or raw wikitext, query-specific summaries, section extraction and internal links, across 50+ languages.
- Article search with relevance ranking
- Several queries run in parallel in one call
- A full article as a plain-text extract or as complete wikitext
- A summary written against the specific question you asked, not the generic lede
- Individual sections extracted and summarised
- Internal links and related articles from an article
- 50+ language codes and matching country codes, resolved automatically
- Input sanitisation and tiered rate limiting on the server side
Nothing to install if you point a client at a deployed instance — it speaks MCP Streamable HTTP and runs stateless, which is correct for serverless. Deploying your own is `vercel` from a clone. Language and country default to `en` and `US`, and both can be set in the environment.
One command — claude mcp add wikipedia --transport http https://your-app.vercel.app/mcp