Search Wikipedia articles and read full content as Markdown via wikipedia-mcp (npx) — 2 tools

wikipedia-mcp v1.0.3 — Search & Read Wikipedia Articles via MCP

Package: wikipedia-mcp (npm) Author: Tim Juenemann Transport: stdio Dependencies: @modelcontextprotocol/sdk, turndown (HTML to Markdown) Credential-free: yes

Tools (2)

Key Observations

1. search returns a snippet, not a list — each search returns ONE best-match article summary (~200-230 chars) plus an article link. No pagination or result count.
2. readArticle returns the FULL article as Markdown — Istanbul article was 432,844 chars. Large articles take 6-7 seconds. Short articles like MCP: 48,421 chars, ~2.7 seconds.
3. pageId works as an alternative to title — both can be used with readArticle. Must provide at least one.
4. Unicode queries supported — searching for Japanese characters correctly returns the Japanese language article.
5. Disambiguation pages returned as-is — searching for Python returns the disambiguation page with links to all meanings; no auto-resolution.
6. Graceful error handling — nonexistent articles return a clear error message; empty search returns No results found.
7. First search is slowest (~1800ms cold start), subsequent searches ~400-600ms.
8. Markdown quality — uses turndown to convert HTML to Markdown. Preserves headings, links, image references, tables.

Gotchas

• No multi-result search — you get exactly 1 result per search call.
• No language parameter — always returns English Wikipedia.
• Large articles are VERY large — 400K+ chars for major cities. No truncation or section-level reading.
• Images are reference-only — Markdown includes image URLs but no actual image data.