wikipedia-mcp-server

Overview

Wikipedia content via the MediaWiki REST API and Action API. Search articles, read summaries or targeted sections, find geotagged pages near a coordinate, and list language editions from any MCP client. Runs as a stdio process, a local Streamable HTTP server, or the public hosted endpoint above.

Tools

Capability reference

wikipedia_search_articles _tool

• Free-text query, ranked by relevance; returns plain-text snippets (HTML stripped), page IDs, and word counts
• Each result carries the article's short description and Wikidata QID (wikibase_item) when it has them, from one follow-up lookup per page of results. That lookup is best-effort: if it fails, the results still come back without the two fields and the notice says so
• Enrichment suggestion carries Wikipedia's spelling correction whenever it has one (einstien → einstein), and a zero-hit first page names it in the notice — re-run with it as query
• limit is 1–50; offset pages further results — enrichment nextOffset signals more remain and is passed back as offset
• Wikipedia serves no result past the 10,000th for a query: an offset at or beyond it fails with offset_too_large, and a page ending on the window carries enrichment truncated naming the matches no offset reaches — narrow the query to bring them into range
• An empty query fails with empty_query; a whitespace-only query is a real search that simply matches nothing
• language selects any Wikipedia edition (default en)
• Best when the exact article title is unknown, or to discover multiple articles on a topic

wikipedia_get_summary _tool

• Returns the REST summary extract — a truncated fragment from the start of the lead, not the whole lead — plus the Wikidata QID (wikibase_item), short description, and thumbnail URL
• url is the canonical article URL, and revision_id / last_modified name the revision the extract was read from — ?oldid=<revision_id> is a permanent link to it
• Superscripts and subscripts in the extract stay distinct from the digits beside them (10²³, H₂O), rendered the same way as on wikipedia_get_article
• latitude / longitude are present for a geotagged article and pass straight to wikipedia_search_nearby, whose inputs carry those names; both are absent otherwise
• For the lead section in full, call wikipedia_get_article with section_index: 0
• page_type discriminates standard / disambiguation / no-extract — on disambiguation, re-query with wikipedia_search_articles for a more specific title
• Redirect pages are followed automatically
• Right tool for most encyclopedic "what is X?" lookups; use wikipedia_get_article for full depth

wikipedia_get_article _tool

• Without section_index: full article with == Section == markers, unless it exceeds WIKIPEDIA_ARTICLE_OVERFLOW_BYTES (default 80,000 bytes) — then returns a section outline (truncated: true) pointing to wikipedia_get_sections plus a targeted section_index read
• With section_index (from wikipedia_get_sections): returns that section plus every nested subsection, each heading above its own body
• section_index: 0 is the lead section — the text above the first heading, returned under the title Introduction
• Both paths render code samples as fenced blocks with indentation intact and formulas as their TeX. Section reads also render data tables as pipe-delimited rows (header row, | --- |, then one line per row; rowspan cells repeated) and infoboxes as label: value lines; a table over 40,000 rendered bytes leaves a [table omitted: N rows] marker. The full-article path carries no tables or infoboxes — upstream extracts strip them — so read the section for those
• Layout-only tables (multi-column lists, succession boxes) keep their content as ordinary text
• Superscripts and subscripts stay distinct from the digits beside them: 10²³, mol⁻¹, H₂O, or ^x / _x where a character has no Unicode form. An abbreviation's superscript stays joined, as the edition writes it in plain text (French XIXe siècle, 1er, Mme)
• Every read returns url (the canonical article URL) and revision_id (the revision the text was read from — ?oldid=<revision_id> is a permanent link to it); a full read, outline included, also returns last_modified, that revision's timestamp. For a redirect, all three name the target article. With WIKIPEDIA_BASE_URL set, a section read omits url, since the mirror's article path is unknown
• Page furniture — maintenance banners, sister-project and library-resource boxes, portal bars, spoken-article notices — is stripped, as are the editor-only preview warnings a section render emits; hatnotes are kept
• Redirect pages are followed automatically

wikipedia_get_sections _tool

• Returns section titles, heading levels, hierarchical numbering (e.g. "2.1"), and section_index values
• The first entry is the lead: index: 0, titled Introduction — Wikipedia's own table of contents starts at the first heading
• section_index is the integer to pass to wikipedia_get_article for a targeted read
• Fails with no_sections on a stub or very short article — read it with wikipedia_get_article instead
• Redirect pages are followed automatically

wikipedia_search_nearby _tool

• Returns geotagged articles sorted ascending by distance, with coordinates, distance_meters, and each article's short description and Wikidata QID (wikibase_item) when it has them
• radius_meters: 10–10,000 (default 1000); limit: 1–500 (default 10) — no pagination past limit, so raise it or sweep narrower radii for full coverage
• Only articles carrying their own coordinate tag (GeoData, set on the article — not the Wikidata item's coordinate) are returned, and that tag places and measures each result. An article with a wrong tag appears where the tag puts it; its description usually gives it away (Palazzo Bernardo Nani — "Palace on the Grand Canal, Venice" — 161 m from the Eiffel Tower)
• Enrichment truncated flags when more articles matched than limit allowed; at limit: 500, Wikipedia's ceiling, a full page reports truncated and the notice points to narrower sweeps rather than a higher limit

wikipedia_get_languages _tool

• Returns each edition's language_code, tool-usable edition_code (can differ, e.g. gsw vs als), article title, and URL
• Pass edition_code — not language_code — as the language parameter on other tools
• editions narrows the answer to the codes asked for, matched against both edition_code and language_code; requested codes with no article come back under missing, and total_languages stays the unfiltered count. A popular article lists hundreds of editions, so the filter is the difference between a 40 KB reply and a 1 KB one
• Fails with no_other_languages when the article has no translations — a filter that matches nothing is a normal response with an empty list, not a failure
• Redirect pages are followed automatically; source_title reports the resolved title

Features

Built on @cyanheads/mcp-ts-core: stdio and Streamable HTTP transports, pluggable auth (none / jwt / oauth), swappable storage (in-memory, filesystem, Supabase, Cloudflare KV/R2/D1), structured logging with optional OpenTelemetry tracing.

Wikipedia-specific:

• Dual API integration — MediaWiki REST API (/api/rest_v1/) for summaries, Action API (/w/api.php) for search, full text, sections, geo search, and language links
• Retry and backoff on every required request, including the transient refusals (search too busy, rate-limited, read-only) the Action API returns inside an HTTP 200 body, with an upstream Retry-After honored and each request's retries bounded at 30 s. The best-effort description lookup on search results gets one short attempt; User-Agent header per Wikimedia API policy
• Every read path renders HTML through one renderer to the same plain-text shape — == Heading == markers, one list item per line, superscripts kept, code fenced: the full article from the Action API's HTML extract, a section from the parser's own HTML for that section, the summary from the REST extract_html. A section read additionally carries data tables, infoboxes, and the lists inside layout tables, none of which the extract carries
• Per-call language parameter on every tool — all Wikipedia language editions accessible in a single session
• Language validation against a live edition registry built from the MediaWiki action=sitematrix endpoint (cached 24h) — catches structurally valid but nonexistent editions before they cause timeouts

Agent-friendly output:

• page_type on summaries discriminates standard / disambiguation / no-extract — no string parsing needed
• wikibase_item (Wikidata QID) on summaries and on search and nearby results enables direct cross-referencing with wikidata-mcp-server
• Article text, snippets, and titles are backslash-escaped on the way into the markdown content[] render, so an article that writes about markup or markdown syntax reads as itself instead of being interpreted by the client; structuredContent carries the same text unescaped. Fenced code blocks and table-row delimiters pass through unescaped, while table cells stay escaped
• section_index on table-of-contents entries links directly to the targeted-read parameter on wikipedia_get_article, index 0 included
• Titles MediaWiki cannot name a page with — < > [ ] { }, the | multi-title separator, percent escapes, magic tildes, relative paths — are refused before any request, with invalid_title; a trailing #fragment is accepted and resolves normally
• Recovery hints on every error type — callers get actionable next steps (e.g., "use wikipedia_search_articles to find the correct title")

Getting started

Public Hosted Instance

A public instance is available at https://wikipedia.caseyjhand.com/mcp — no installation required. Point any MCP client at it via Streamable HTTP:

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

Self-Hosted / Local

Add the following to your MCP client configuration file.

{
  "mcpServers": {
    "wikipedia-mcp-server": {
      "type": "stdio",
      "command": "bunx",
      "args": ["@cyanheads/wikipedia-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with npx (no Bun required):

{
  "mcpServers": {
    "wikipedia-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@cyanheads/wikipedia-mcp-server@latest"],
      "env": {
        "MCP_TRANSPORT_TYPE": "stdio",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}

Or with Docker:

{
  "mcpServers": {
    "wikipedia-mcp-server": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "MCP_TRANSPORT_TYPE=stdio",
        "ghcr.io/cyanheads/wikipedia-mcp-server:latest"
      ]
    }
  }
}

For Streamable HTTP, set the transport and start the server:

MCP_TRANSPORT_TYPE=http MCP_HTTP_PORT=3010 bun run start:http
# Server listens at http://localhost:3010/mcp

Prerequisites

• Bun v1.3.0 or higher (or Node.js v24+).
• No API keys required — Wikipedia's API is public.

Installation

1. Clone the repository:

git clone https://github.com/cyanheads/wikipedia-mcp-server.git

2. Navigate into the directory:

cd wikipedia-mcp-server

3. Install dependencies:

bun install

4. Configure environment (optional):

cp .env.example .env
# edit .env if you want to customize WIKIPEDIA_USER_AGENT or logging

Configuration

See .env.example for the full list of optional overrides.

Running the server

Local development

• Build and run:

  # One-time build
  bun run rebuild
  
  # Run the built server
  bun run start:stdio
  # or
  bun run start:http
  

• Run checks and tests:

  bun run devcheck   # Lint, format, typecheck, security
  bun run test       # Vitest test suite
  bun run lint:mcp   # Validate MCP definitions against spec
  

Docker

docker build -t wikipedia-mcp-server .
docker run --rm -p 3010:3010 wikipedia-mcp-server

The Dockerfile defaults to HTTP transport, stateless session mode, and logs to /var/log/wikipedia-mcp-server. OpenTelemetry peer dependencies are installed by default — build with --build-arg OTEL_ENABLED=false to omit them.

Project structure

Development guide

See CLAUDE.md for development guidelines and architectural rules. The short version:

• Handlers throw, framework catches — no try/catch in tool logic
• Use ctx.log for request-scoped logging, ctx.state for tenant-scoped storage
• Register new tools in src/mcp-server/tools/definitions/index.ts
• Wrap external API calls: validate raw → normalize to domain type → return output schema; never fabricate missing fields

Contributing

Issues are welcome. Run checks and tests before submitting:

bun run devcheck
bun run test

License

Apache-2.0 — see LICENSE for details.