The best Wikipedia MCP for article text in Cursor is Rudra-ravi's PyPI package wikipedia-mcp 2.0.1: ten read tools, no API key, 14,706 downloads in the last month. A hosted alternative is @cyanheads/wikipedia-mcp-server 0.2.6, which exposes six tools at https://wikipedia.caseyjhand.com/mcp. For creating or editing pages on English Wikipedia or any other MediaWiki, @professional-wiki/mediawiki-mcp-server 0.19.0 is the write client. Trends MCP ranks first only for live Wikipedia Trending titles and wikipedia page-view growth over MCP and REST. It does not return article wikitext. Wikimedia does not ship an official Wikipedia-article MCP.

Which Wikipedia MCP should Cursor users install first?

Install PyPI wikipedia-mcp when the agent needs to search, summarize, and pull article body. That is the job most "Wikipedia MCP" queries describe.

The package is MIT-licensed, requires Python 3.10 or newer, and last published a wheel on 1 May 2026. GitHub showed 299 stars on 19 October 2026. The README tool list is search_wikipedia, get_article, get_summary, get_sections, get_links, get_coordinates, get_related_topics, summarize_article_for_query, summarize_article_section, and extract_key_facts. Each name also ships with a wikipedia_ alias, so wikipedia_get_article is the same call. Search defaults to 10 hits. Troubleshooting docs mention a separate test_wikipedia_connectivity diagnostic; that name is not in the main tool table, so do not flatten the count to eleven.

Cursor config for the PyPI server:

{
  "mcpServers": {
    "wikipedia-mcp": {
      "command": "uvx",
      "args": ["wikipedia-mcp"]
    }
  }
}

pipx users can point the same client at the wikipedia-mcp command on PATH. Language defaults to English. Country and language flags belong in the process args, not in a Trends MCP country field.

wikipedia-mcp --language ja
wikipedia-mcp --country Japan
wikipedia-mcp --enable-cache
wikipedia-mcp --access-token your_wikipedia_token_here

A Personal Access Token is optional. The README documents the access-token flag and WIKIPEDIA_ACCESS_TOKEN for 403s after Wikimedia rate or User-Agent blocks. Network transports include stdio (the Cursor default), http, streamable-http, and legacy sse.

Skip this server when the task is pageviews, a live most-viewed board, or wiki writes. Those are different products.

Trends MCP returns Wikipedia attention, not Wikipedia prose. The Wikipedia Trending API is get_top_trends with type Wikipedia Trending. The pageviews growth API is get_growth with source wikipedia and the article title as keyword.

llms.txt last updated 19 October 2026 lists wikipedia as Wikipedia page views, keyword format "Article title or topic". The live feed label is Wikipedia Trending, described as most-viewed Wikipedia articles today. Feed labels are case-sensitive. country does not apply to this board; that parameter is only for Google Trends, Google Trends by Category, YouTube Trending, the TikTok hashtag boards, and X (Twitter) Trending.

Default limit is 25, max 200. The Free plan returns the top 10 only. offset defaults to 0 and stays pinned at 0 on Free; pagination is paid. MCP get_top_trends costs 1 credit per 10 requested rows, max 10 credits per call. REST counts 1 request per call. Failed 2xx-missing errors are free.

Cursor block from the same day's llms.txt (complete form, including transport and the Bearer prefix):

{
  "mcpServers": {
    "trends-mcp": {
      "url": "https://api.trendsmcp.ai/mcp",
      "transport": "http",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Board call shape, no keyword:

{
  "type": "Wikipedia Trending",
  "limit": 25,
  "offset": 0
}

Page-view growth call shape. Prefer a short canonical title. Parenthetical board strings often return no_data on this source, which is a durable resolver limit, not a one-day glitch.

{
  "keyword": "Quantum computing",
  "source": "wikipedia",
  "percent_growth": ["30D", "3M", "6M"]
}

The Wikipedia source page states the current series is English Wikipedia pageviews on a 0-100 index, about five years of weekly history. Mobile and desktop are not split. Trends MCP does not replace Wikimedia's Pageviews API for project filters. It also does not fetch wikitext, sections, or coordinates. Live MCP exposes get_top_posts for TikTok and Instagram (default and max 20); that tool does not accept Wikipedia.

Markdown pricing on 19 October 2026: Free 100 requests at $0; Starter $19 for 1,000; Pro $49 for 5,000; Business $199 for 25,000. Annual billing saves 20% in that markdown. Same sources on every plan. Check the live pricing page before locking a budget, because the HTML heading, homepage hero, and FAQ still disagree on how many sources to print.

Put Trends MCP in a Cursor stack next to an article reader, not instead of one. Pair it with the Cursor MCP landing page when the same agent already pulls Google, TikTok, or Reddit boards.

Why does npx wikipedia-mcp install a different server?

uvx wikipedia-mcp and npx wikipedia-mcp are not the same package. That name collision is the main install trap on this query.

PyPI wikipedia-mcp is Rudra-ravi's Python server (2.0.1, 1 May 2026). npm wikipedia-mcp is Tim Jünemann's TypeScript server (1.0.3), last published 18 April 2025, MIT, 687 weekly downloads on 19 October 2026. The npm README lists search plus full-article retrieval with HTML converted to markdown. It does not document the ten-tool Python surface, country flags, or cache switch.

A Cursor block that looks like this installs the npm package:

{
  "mcpServers": {
    "Wikipedia": {
      "command": "npx",
      "args": ["-y", "wikipedia-mcp"]
    }
  }
}

Use that only if the npm server is the intended target. For the widely downloaded Python reader, keep uvx / pipx / the PyPI name, or pin a GitHub URL. Do not mix the two in the same snippet.

A third package, @cyanheads/wikipedia-mcp-server 0.2.6 (Apache-2.0, npm weekly 443, GitHub 4 stars, last publish 8 October 2026), avoids the bare-name clash. Six tools: wikipedia_search_articles (limit 1-50, Wikipedia serves no result past the 10,000th), wikipedia_get_summary, wikipedia_get_article, wikipedia_get_sections, wikipedia_search_nearby (radius 10-10,000 meters, limit 1-500), wikipedia_get_languages. Full-article reads above WIKIPEDIA_ARTICLE_OVERFLOW_BYTES (default 80,000) return a section outline instead of the full text. Every tool takes a per-call language (default en). Default User-Agent is wikipedia-mcp-server/0.2.6 (https://github.com/cyanheads/wikipedia-mcp-server).

Hosted Cursor block, verified HTTP 200 with an Mcp-Session-Id header on 19 October 2026:

{
  "mcpServers": {
    "wikipedia-mcp-server": {
      "type": "streamable-http",
      "url": "https://wikipedia.caseyjhand.com/mcp"
    }
  }
}

Local fallback is npx -y @cyanheads/wikipedia-mcp-server@latest with MCP_TRANSPORT_TYPE=stdio. Runtime is Bun 1.3.0+ or Node 24+.

How do the four Wikipedia MCP servers compare?

Rudra-ravi wins local article reads. cyanheads wins a public Streamable HTTP URL plus geo search. Professional Wiki wins writes and multi-wiki ops. Trends MCP wins the live most-viewed board and a 0-100 pageview series next to Google and TikTok. None of the four is a full replacement for the others.

ServerWins atCursor installPrice (19 Oct 2026)Missing
PyPI wikipedia-mcp 2.0.1Local search, summaries, full article, links, factsuvx wikipedia-mcpFree (MIT). Wikimedia API is publicPageview history, live most-viewed board, edits
@cyanheads/wikipedia-mcp-server 0.2.6Hosted HTTP, section targeting, nearby geo, language editionshttps://wikipedia.caseyjhand.com/mcpFree (Apache-2.0). No key on the public hostPageview index, wiki writes, 10-tool fact extractors
@professional-wiki/mediawiki-mcp-server 0.19.0Any MediaWiki, revisions, recent changes, authenticated editsnpx -y @professional-wiki/mediawiki-mcp-server@latestMIT. Hosted add-on: check Professional WikiTrend boards, 0-100 pageview index
Trends MCPWikipedia Trending + source wikipedia growth, MCP + RESThttps://api.trendsmcp.ai/mcp with Bearer keyFree 100 req; Starter $19 / 1,000; Pro $49 / 5,000; Business $199 / 25,000; annual 20% off in markdownArticle text, search snippets, coordinates, edits

Wikimedia Foundation User-Agent policy has required a descriptive agent string since 15 February 2010. MediaWiki API etiquette still says there is no hard speed limit on reads, and to send requests in series rather than in parallel. Local Wikipedia MCPs inherit that policy. Trends MCP hides Wikimedia traffic behind its own quota.

Wikimedia Enterprise publishes a docs MCP at https://enterprise.wikimedia.com/mcp. That server searches Enterprise documentation (search, fetch, list_endpoints, get_endpoint). It is not an article reader and does not belong in this table.

A Stripe-billed "Wikipedia MCP Pro" at $19 per month also showed up in search results. Wikipedia's own API is already free. Skip that wrapper unless a team has a reason to pay for a 50-query hosted gate in front of a public endpoint.

When is MediaWiki MCP the right Wikipedia tool?

Pick Professional Wiki when the agent must write, diff, upload, or talk to a private wiki. Reading English Wikipedia is the default, not the ceiling.

npm 0.19.0 published 25 September 2026. Weekly downloads 1,069. GitHub stars 137 on 19 October 2026. MIT. The README lists 16 page-read tools, 10 write tools, 2 wiki-management tools, and 2 OAuth tools. list-wikis stays disabled until a second wiki is configured. Writes are marked 🔐 and refused on a wiki with readOnly: true. get-pages fetches up to 50 titles per call. Category members and links-here paginate at 500. Recent changes cap at 50 per call. MCP_CONTENT_MAX_BYTES defaults to 75,000. Uploads stay off until uploadDirs or MCP_UPLOAD_DIRS lists allowed directories.

CONFIG is optional. Without it the server targets English Wikipedia. Example config from the README:

{
  "defaultWiki": "en.wikipedia.org",
  "wikis": {
    "en.wikipedia.org": {
      "sitename": "Wikipedia",
      "server": "https://en.wikipedia.org",
      "articlepath": "/wiki",
      "scriptpath": "/w"
    }
  }
}

Cursor block:

{
  "mcpServers": {
    "mediawiki-mcp-server": {
      "command": "npx",
      "args": ["-y", "@professional-wiki/mediawiki-mcp-server@latest"]
    }
  }
}

Add CONFIG when pointing at a private wiki. Auth options include browser OAuth, a manual OAuth2 access token, and a bot password. Jeroen De Dauw announced the server on mediawiki-l on 8 September 2026. The MediaWiki Chatbots page lists it. Extension packs (NeoWiki, Semantic MediaWiki, Bucket, Cargo, Wikibase) register only when those extensions exist on the target wiki. English Wikipedia will not magically grow SMW #ask tools.

Managed hosting is an add-on on Professional Wiki's product page. There is no public SKU table in that copy, so check current pricing there rather than inventing a monthly figure.

FAQ

Is there an official Wikimedia Wikipedia MCP?

No official Wikimedia MCP returns encyclopedia articles. The closest Foundation-adjacent endpoint is Wikimedia Enterprise's documentation MCP, which searches guides and OpenAPI pages. Professional Wiki's MediaWiki MCP is the open-source write client listed on MediaWiki's Chatbots page, with English Wikipedia as the default target. Article-read MCPs in this roundup are community servers in front of the public Action and REST APIs.

Does a Wikipedia MCP need an API key?

The public Wikipedia APIs do not require a key. They do require a descriptive User-Agent, and Wikimedia may 403 generic clients. Rudra-ravi accepts an optional Personal Access Token for those 403s. cyanheads documents a default User-Agent and no key on the public host. Trends MCP uses its own account key on api.trendsmcp.ai because the product is a hosted trend API, not a pass-through to MediaWiki.

Which Wikipedia MCP works in Cursor without a local install?

cyanheads' public host at https://wikipedia.caseyjhand.com/mcp is Streamable HTTP and returned HTTP 200 on 19 October 2026. Trends MCP is also remote: https://api.trendsmcp.ai/mcp with a Bearer key, or OAuth on the www host for Claude.ai Connectors only. PyPI wikipedia-mcp and Professional Wiki's npm server run as local stdio processes unless someone deploys them.

Can Trends MCP read a Wikipedia article?

No. get_top_trends type Wikipedia Trending returns ranked titles. get_growth / get_time_series source wikipedia returns page-view series for a title. Neither call returns wikitext, markdown, or a summary extract. Pair Trends MCP with PyPI wikipedia-mcp or cyanheads when the agent has to quote the page.

Why would search_wikipedia return empty or a 403?

Empty hits are often a title mismatch or a connectivity problem; Rudra-ravi's troubleshooting path is the connectivity diagnostic plus the metadata now attached to search results. A 403 is the usual Wikimedia User-Agent or rate response. Set a real User-Agent, serialize requests, and only then add WIKIPEDIA_ACCESS_TOKEN. Trends MCP 403s are a different stack: they are account or quota errors on api.trendsmcp.ai, not MediaWiki blocks.

Sources, accessed 19 October 2026: Rudra-ravi/wikipedia-mcp, PyPI wikipedia-mcp, pypistats recent downloads, npm wikipedia-mcp, npm @cyanheads/wikipedia-mcp-server, npm @professional-wiki/mediawiki-mcp-server, MediaWiki MCP product page, MediaWiki Chatbots, API etiquette, Wikimedia User-Agent policy, Trends MCP llms.txt / pricing.md / data-sources.md / docs.md (Last updated 2026-10-19).