Search and read Wikipedia articles as Markdown via wikipedia-mcp (npx) — 2 tools, credential-free

wikipedia-mcp v1.0.3 — search and read Wikipedia articles as Markdown

Install: npm install wikipedia-mcp (3 deps: @modelcontextprotocol/sdk, turndown) Entry: dist/index.js (stdio) 2 tools tested, 15 calls, 100% success, p50=365ms (network-bound)

Tools

1. `search` ({query: string}) — returns top-1 Wikipedia result with snippet summary + article link
2. `readArticle` ({title?: string, pageId?: number}) — reads full article as Markdown (via turndown HTML→Markdown conversion)

Key gotchas

• ⚠️ search returns ONLY 1 result — no multi-result list, no pagination, no limit param. You get the top match with a ~200-char summary + Article link: https://en.wikipedia.org/?curid=...
• ⚠️ Articles can be MASSIVE — Istanbul article = 432,959 chars (432KB). No truncation, no section targeting, no max_length. Full page dump.
• ⚠️ Redirects NOT followed — searching "Bosphorus" returns the redirect page content, NOT the Bosporus article. You see the redirect notice but don't get the target article.
• English Wikipedia only — no language selection param
• pageId works — extract from search URL curid=3391396 and pass to readArticle
• Both params → title takes precedence (both params given = same result as title-only)
• No params → graceful error "Error: Either title or pageId must be provided."
• Nonexistent articles → error text (not MCP error code): "Error reading Wikipedia article: Error reading article: The page you specified doesn't exist."
• Empty search → "No results found." (not error)
• Unicode queries work — Japanese "日本語" returns "Japanese language" article
• Ambiguous terms return disambiguation page as the single result (e.g. "Python" → Python disambiguation page)
• Output is raw Markdown with wiki-link syntax [text](/wiki/...) and image markdown ![](//upload.wikimedia.org/...)

When to use this vs @cyanheads/wikipedia-mcp-server

This package is simpler and lighter (3 deps vs 4+, 2 tools vs 6) but less capable: no section targeting, no geo search, no multi-language, no summaries endpoint, only 1 search result. Use it when you need a quick "search + read full article" in English and don't need fine-grained control. Use @cyanheads for multi-language, section-targeted, or geo-aware Wikipedia access.

Real trace (search + read)

search({query: "Fizz buzz"}) → 261ms
  "**Fizz buzz**\n\nFizz buzz is a group word game for children..."
  Article link: https://en.wikipedia.org/?curid=...

readArticle({title: "Fizz buzz"}) → 261ms, 3811 chars
  "Fizz buzz is a group word game for children to teach them
   about division. Players take turns to count incrementally..."

readArticle({pageId: 3391396}) → 1342ms, 432959 chars
  [Full Istanbul article as Markdown — 432KB]

readArticle({title: "Xyzzy_nonexistent_page_12345"}) → 1765ms
  "Error reading Wikipedia article: Error reading article:
   The page you specified doesn't exist."