Wiki Knowledge MCP Server

Wiki Knowledge MCP Server gives Claude, ChatGPT, Cursor and any other MCP client 14 read-only tools over content derived from Wikipedia in 348 language editions, the Wikidata knowledge graph and English definitions from Wiktionary. It runs on Apify in Standby mode, needs no key or account on the source side, and every answer carries its licence and a link to the page it came from. Wikipedia, Wikidata and Wiktionary are trademarks of the Wikimedia Foundation. This server is not endorsed by or affiliated with the Wikimedia Foundation.

Ask Claude which articles describe places within a kilometre of the Eiffel Tower. It reads the tower's coordinates with wiki_get_article_summary, then calls wiki_find_articles_near, which returned 89 geotagged articles sorted by distance in a platform test, each with its link. Ask for the films Christopher Nolan directed and wiki_query_entities builds the knowledge graph query from ids, so nobody writes SPARQL.

What this server exposes

This server reads three public knowledge sources through the Wikimedia Foundation APIs and one community dictionary service: Wikipedia in its 348 language editions (article search, introductions, full text section by section, language versions, articles near a point), Wikidata (item and property search, facts with readable labels, lookups by external identifier, and two query builders that list or rank items from ids only, never from raw SPARQL), and English definitions from Wiktionary, served by the Free Dictionary API with the English Wiktionary as a fallback. It exposes 12 read-only tools with the wiki_ prefix, plus the search and fetch pair that ChatGPT deep research expects over English Wikipedia. No upstream needs a key or an account, and the server stores nothing about its callers beyond a shared response cache. Every answer says where the content comes from, under which licence, and links to the page and its history.

Tools

Tool	What it answers	Billed event
wiki_search_articles	Search encyclopedia articles	tool-read
wiki_get_article_summary	Get an article summary	tool-read
wiki_get_article_sections	List the sections of an article	tool-read
wiki_read_article	Read an article	tool-read
wiki_get_article_languages	List the language versions of an article	tool-read
wiki_find_articles_near	Find articles about places near a point	tool-read
wiki_search_entities	Find knowledge graph ids	tool-read
wiki_get_entity	Get the facts of a knowledge graph item	tool-read
wiki_resolve_external_id	Find the item behind an external identifier	tool-read
wiki_query_entities	List knowledge graph items by class and relations	tool-action
wiki_rank_entities	Rank knowledge graph items by a value	tool-action
wiki_define_word	Define an English word	tool-read
search	Search	tool-read
fetch	Fetch	tool-read

wiki_search_articles: Search encyclopedia articles

Use this when you need to find Wikipedia articles about a topic, person, place or event, in any language edition, before reading them. Returns titles, plain-text snippets and article URLs ranked by relevance, plus a "did you mean" suggestion when Wikipedia offers one. Do not use for Wikidata items or ids (use wiki_search_entities) or when you already know the exact title (use wiki_get_article_summary). Parameters: query (string, e.g. "causes of the french revolution"), language (edition code, default "en", e.g. "fr"), limit (1-50, default 10), offset (0-9,999, default 0), response_format ("concise" default, "detailed" adds page id, word count and last edit date).

Parameter	Type	Required	Example
query	string	yes	"causes of the french revolution"
language	string	no	"en"
limit	integer	no	10
offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { language, query, total_hits, suggestion, results[], pagination, attribution }. Upstream: Wikipedia Action API. Cache: hourly (1 hour). Read-only.

wiki_get_article_summary: Get an article summary

Use this when you need a quick answer to "what or who is X": the introduction of one Wikipedia article in plain text, its short description, its Wikidata id and coordinates when it has them. Redirects are followed. The answer names the article, its URL and its history page for the CC BY-SA attribution. Do not use for the full text or a given section (use wiki_read_article) or when you do not know the exact title (use wiki_search_articles first). Parameters: title (article title or article URL, e.g. "Marie Curie"), language (edition code, default "en"), response_format ("concise" default, "detailed" adds page id, revision, last edit date and thumbnail image).

Parameter	Type	Required	Example
title	string	yes	"Marie Curie"
language	string	no	"en"
response_format	"concise" or "detailed"	no	"concise"

Returns { language, title, description, type, is_disambiguation, extract, url, history_url, wikidata_id, coordinates, page_id, revision, last_edited, thumbnail, notices[], attribution }. Upstream: Wikipedia REST API. Cache: daily (1 day). Read-only.

wiki_get_article_sections: List the sections of an article

Use this when you need the table of contents of one Wikipedia article: every section with its index, heading level and size in characters, before reading only the part you need. Index 0 is the introduction. Redirects are followed. Do not use for the text itself (use wiki_read_article with a section title or index) or for a short answer (use wiki_get_article_summary). Parameters: title (article title or article URL, e.g. "Photosynthesis"), language (edition code, default "en").

Parameter	Type	Required	Example
title	string	yes	"Marie Curie"
language	string	no	"en"

Returns { language, title, url, history_url, redirected_from, wikidata_id, is_disambiguation, total_characters, section_count, sections[], notices[], attribution }. Upstream: Wikipedia Action API. Cache: daily (1 day). Read-only.

wiki_read_article: Read an article

Use this when you need the actual text of a Wikipedia article, the whole article page by page or one section with its subsections, as plain text (tables, images and references removed). Do not use for a short definition (use wiki_get_article_summary) or to find which sections exist (use wiki_get_article_sections). Parameters: title (article title or URL, e.g. "Photosynthesis"), language (default "en"), section (optional section title or index, e.g. "History" or "3"), limit (sections per call, 1-20, default 5), offset (section position to start from, default 0), max_characters (text budget per call, 1,000-20,000, default 8,000), char_offset (continue a cut section, default 0), response_format ("concise" default, "detailed" adds levels and sizes).

Parameter	Type	Required	Example
title	string	yes	"Marie Curie"
language	string	no	"en"
section	string	no	"History"
limit	integer	no	5
offset	integer	no	0
max_characters	integer	no	8000
char_offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { language, title, url, history_url, redirected_from, selected_section, sections[], notices[], pagination, attribution }. Upstream: Wikipedia Action API. Cache: daily (1 day). Read-only.

wiki_get_article_languages: List the language versions of an article

Use this when you need to know in which Wikipedia language editions an article exists and what its title and URL are in each one, for example to read the French or Japanese version of an English article. Do not use for translating arbitrary words (use wiki_define_word for English definitions) or for Wikidata labels (use wiki_get_entity with a language). Parameters: title (article title or URL, e.g. "Eiffel Tower"), language (edition of that title, default "en"), limit (1-100, default 50), offset (default 0), response_format ("concise" default, "detailed" adds the native language name).

Parameter	Type	Required	Example
title	string	yes	"Marie Curie"
language	string	no	"en"
limit	integer	no	50
offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { language, title, url, languages[], pagination, attribution }. Upstream: Wikipedia Action API. Cache: daily (1 day). Read-only.

wiki_find_articles_near: Find articles about places near a point

Use this when you need the Wikipedia articles about landmarks, buildings, streets, towns or sites located near a latitude and longitude, sorted by distance, for example to describe the surroundings of an address. Covers geotagged articles only, at most 100 within the radius. Do not use for geocoding a place name (search it with wiki_search_articles and read its coordinates with wiki_get_article_summary). Parameters: latitude (-90 to 90, e.g. 48.8584), longitude (-180 to 180, e.g. 2.2945), radius_m (10-10,000 metres, default 1,000), language (default "en"), limit (1-100, default 20), offset (default 0), response_format ("concise" default, "detailed" adds coordinates, page id and place type).

Parameter	Type	Required	Example
latitude	number	yes	48.8584
longitude	number	yes	2.2945
radius_m	integer	no	1000
language	string	no	"en"
limit	integer	no	20
offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { language, center, radius_m, articles[], pagination, attribution }. Upstream: Wikipedia Action API. Cache: daily (1 day). Read-only.

wiki_search_entities: Find knowledge graph ids

Use this when you need the Wikidata id of a thing, person, place, concept (items, Q...) or of a relation (properties, P...) from its name or an alias, before calling wiki_get_entity, wiki_query_entities or wiki_rank_entities. Matches labels and aliases, not full text. Do not use for reading facts (use wiki_get_entity) or for finding encyclopedia articles (use wiki_search_articles). Parameters: query (string, e.g. "Douglas Adams" or "population"), language (language of the labels to match and show, default "en"), entity_type ("item" default, or "property"), limit (1-50, default 10), offset (default 0), response_format ("concise" default, "detailed" adds the matched text and aliases).

Parameter	Type	Required	Example
query	string	yes	"Douglas Adams"
language	string	no	"en"
entity_type	"item" or "property"	no	"item"
limit	integer	no	10
offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { language, query, entity_type, results[], pagination, attribution }. Upstream: Wikidata API. Cache: daily (1 day). Read-only.

wiki_get_entity: Get the facts of a knowledge graph item

Use this when you need what Wikidata states about one item or property: label, description, aliases, Wikipedia link and its statements (instance of, country, dates, population, occupation, identifiers...) with readable labels. Key facts come first. Only best-rank values are shown. For people, special-category data (religion, ethnicity, health, sexual orientation, political party or ideology, convictions) is withheld. Do not use for finding an id from a name (use wiki_search_entities) or for lists across many items (use wiki_query_entities). Parameters: id (e.g. "Q42", "P31" or a Wikidata URL), language (labels, default "en"), limit (properties per page, 1-30, default 20), offset (default 0), response_format ("concise" default with up to 3 values per property, "detailed" up to 6 with date qualifiers).

Parameter	Type	Required	Example
id	string	yes	"Q42"
language	string	no	"en"
limit	integer	no	20
offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { id, type, label, description, aliases[], url, wikipedia, sitelink_count, property_count, special_categories_withheld, statements[], notices[], pagination, attribution }. Upstream: Wikidata API. Cache: daily (1 day). Read-only.

wiki_resolve_external_id: Find the item behind an external identifier

Use this when you have an identifier from another catalogue (IMDb, DOI, ORCID, VIAF, GND, Library of Congress, ISO country, language or currency code, IATA or ICAO airport code, LEI, MusicBrainz, GeoNames, OpenStreetMap relation, InChIKey, PubChem, CAS) and need the matching Wikidata item and Wikipedia article. The value is normalised (case, prefixes, the catalogue URL forms, ORCID check digit) before an exact match. Do not use for names (use wiki_search_entities). Parameters: id_type (one of the listed schemes, e.g. "imdb"), value (e.g. "tt0111161"), language (labels and Wikipedia link, default "en"), response_format ("concise" default, "detailed" adds the Wikidata property used and cache information).

Parameter	Type	Required	Example
id_type	"imdb" or "doi" or "orcid" or "viaf" or "gnd" or "library_of_congress" or "iso_3166_1_alpha2" or "iso_3166_1_alpha3" or "iso_639_1" or "iso_639_3" or "iso_4217" or "iata_airport" or "icao_airport" or "lei" or "musicbrainz_artist" or "geonames" or "openstreetmap_relation" or "inchikey" or "pubchem_cid" or "cas_number"	yes	"imdb"
value	string	yes	"tt0111161"
language	string	no	"en"
response_format	"concise" or "detailed"	no	"concise"

Returns { id_type, value, property_id, property, matches[], notices[], attribution }. Upstream: Wikidata API. Cache: daily (1 day). Read-only.

wiki_query_entities: List knowledge graph items by class and relations

Use this when you need a list of Wikidata items that share a class and relations, optionally within a date range: films directed by someone, people of an occupation born in a country between two years, museums in a city, books by an author. Builds a safe query from ids only (no raw SPARQL). Ordered by popularity (number of Wikimedia sites about the item) or by date. Do not use for ranking by a number such as population (use wiki_rank_entities) or for one item (use wiki_get_entity). Parameters: instance_of (class id, e.g. "Q11424" film), include_subclasses (default false), filters (0-3 of {property_id, value_id}, e.g. P57 director = Q25191), date_property (e.g. "P577" publication date), date_from and date_to (years), order_by ("popularity" default, "date_ascending", "date_descending"), language (labels, default "en"), limit (1-100, default 20), offset (0-5,000), response_format ("concise" default, "detailed" adds descriptions and the query summary).

Parameter	Type	Required	Example
instance_of	string	no	"Q515"
include_subclasses	boolean	no	false
filters	array of object	no	[{"property_id":"P17","value_id":"Q142"}]
date_property	string	no	"P577"
date_from	integer	no	1990
date_to	integer	no	1999
order_by	"popularity" or "date_ascending" or "date_descending"	no	"popularity"
language	string	no	"en"
limit	integer	no	20
offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { results[], query_summary, notices[], pagination, attribution }. Upstream: Wikidata Query Service. Cache: daily (1 day). Read-only.

wiki_rank_entities: Rank knowledge graph items by a value

Use this when you need the top or bottom items of a class by a numeric or date value from Wikidata: largest cities of a country by population, the most populous countries, longest rivers, oldest universities by inception. Builds a safe query from ids only (no raw SPARQL) and uses each item's best-rank value. A very large class with a broad filter (the mountains of a whole country, for example) can exceed the 30 s query limit: the TIMEOUT hint then says what to narrow. Do not use for lists without a value to rank on (use wiki_query_entities). Parameters: instance_of (class id, e.g. "Q515" city), rank_by_property (e.g. "P1082" population, "P2044" elevation, "P2046" area, "P571" inception), direction ("descending" default, or "ascending"), include_subclasses (default false), filters (0-3 of {property_id, value_id}, e.g. P17 = Q142 France), language (labels, default "en"), limit (1-50, default 10), offset (0-1,000), response_format ("concise" default, "detailed" adds descriptions and the query summary).

Parameter	Type	Required	Example
instance_of	string	yes	"Q515"
rank_by_property	string	yes	"P1082"
direction	"descending" or "ascending"	no	"descending"
include_subclasses	boolean	no	false
filters	array of object	no	[{"property_id":"P17","value_id":"Q142"}]
language	string	no	"en"
limit	integer	no	10
offset	integer	no	0
response_format	"concise" or "detailed"	no	"concise"

Returns { rank_by_property, direction, results[], query_summary, notices[], pagination, attribution }. Upstream: Wikidata Query Service. Cache: daily (1 day). Read-only.

wiki_define_word: Define an English word

Use this when you need the meaning of an English word or short expression: pronunciation, parts of speech, definitions, examples, synonyms and antonyms, from Wiktionary. Answers come from the Free Dictionary API, or from the English Wiktionary when that service is down or has no entry. Do not use for other languages, for encyclopedia topics (use wiki_get_article_summary) or for translations (use wiki_get_article_languages). Parameters: word (e.g. "serendipity", case-insensitive), response_format ("concise" default with up to 3 definitions per part of speech, "detailed" up to 10 with examples, synonyms per definition and licensed pronunciation audio).

Parameter	Type	Required	Example
word	string	yes	"serendipity"
response_format	"concise" or "detailed"	no	"concise"

Returns { word, source, phonetic, phonetics[], meanings[], source_urls[], attribution }. Upstream: Free Dictionary API, Wiktionary REST API. Cache: reference (30 days). Read-only.

search: Search

Use this when ChatGPT needs to find English Wikipedia articles on a topic for research. Returns up to 8 articles with ids for fetch and their URLs. Do not use for other language editions or for paging through results (use wiki_search_articles). Parameter query (string), e.g. "history of the printing press".

Parameter	Type	Required	Example
query	string	yes	"history of the printing press"

Returns { results[], notices[] }. Upstream: Wikipedia Action API, Wikidata API. Cache: daily (1 day). Read-only.

fetch: Fetch

Use this when ChatGPT needs the full text of one article returned by search, as plain text with its CC BY-SA attribution (at most 30,000 characters), or the facts of one Wikidata item. Do not use for ids that did not come from search, other than a Wikidata id. Parameter id, e.g. "wikipedia:en:Printing_press" or "wikidata:Q42".

Parameter	Type	Required	Example
id	string	yes	"wikipedia:en:Printing_press"

Returns { id, title, text, url, metadata, notices[] }. Upstream: Wikipedia Action API, Wikidata API. Cache: daily (1 day). Read-only.

Connect

Endpoint (Streamable HTTP): https://mrbridge--wiki-knowledge-mcp.apify.actor/mcp

Every client authenticates with your Apify API token, found in Apify Console > Settings > API & Integrations. Clients that support custom headers send it as Authorization: Bearer <token>. The others use the ?token= URL variant, which must be treated as a secret: keep it out of commits, screenshots and shared logs.

Claude Code

$claude mcp add --transport http wiki-knowledge "https://mrbridge--wiki-knowledge-mcp.apify.actor/mcp" --header "Authorization: Bearer ${APIFY_TOKEN}"

Cursor, VS Code and other JSON configurations

{
  "mcpServers": {
    "wiki-knowledge": {
      "type": "http",
      "url": "https://mrbridge--wiki-knowledge-mcp.apify.actor/mcp",
      "headers": {
        "Authorization": "Bearer ${APIFY_TOKEN}"
      }
    }
  }
}

claude.ai and Claude Desktop

Add a custom connector with the URL https://mrbridge--wiki-knowledge-mcp.apify.actor/mcp?token=YOUR_APIFY_TOKEN and no authentication. Prefilled link (replace the token placeholder before saving): add Wiki Knowledge MCP Server to Claude.

ChatGPT (developer mode)

Settings > Apps > Create, URL https://mrbridge--wiki-knowledge-mcp.apify.actor/mcp?token=YOUR_APIFY_TOKEN, authentication "No authentication". ChatGPT snapshots the tool list: after a server update, an admin has to click Refresh on the app.

OpenAI Responses API

{
  "type": "mcp",
  "server_label": "wiki_knowledge",
  "server_url": "https://mrbridge--wiki-knowledge-mcp.apify.actor/mcp",
  "headers": {
    "Authorization": "Bearer YOUR_APIFY_TOKEN"
  },
  "require_approval": "never"
}

Add allowed_tools to restrict the tool list. Tool calls must finish within about 60 seconds on this API, and every tool of this server answers in under 50 seconds.

Anthropic Messages API

{
  "mcp_servers": [
    {
      "type": "url",
      "url": "https://mrbridge--wiki-knowledge-mcp.apify.actor/mcp",
      "name": "wiki-knowledge",
      "authorization_token": "YOUR_APIFY_TOKEN"
    }
  ],
  "tools": [
    {
      "type": "mcp_toolset",
      "mcp_server_name": "wiki-knowledge"
    }
  ]
}

Send the beta header anthropic-beta: mcp-client-2025-11-20.

One-click OAuth alternative

https://mcp.apify.com/?tools=mrbridge/wiki-knowledge-mcp connects through Apify OAuth without a token. It exposes the Actor as a single tool rather than the granular tools listed above.

Cost control

Each Standby run has a spending limit (maxTotalChargeUsd). When a call uses the last of it, the result is still returned, and later calls in that run answer CHARGE_LIMIT_REACHED until the limit is raised.

Why use Wiki Knowledge MCP Server

A language model that answers from memory often gets dates, populations and spellings wrong. This server lets it check them at the source and cite the page. Here is what it adds over a raw API call.

• Long articles in pieces. wiki_get_article_sections lists the table of contents, then wiki_read_article returns one section or a page of sections within a character budget (8,000 by default, up to 20,000). The live article on photosynthesis holds 49,740 characters in 33 sections, and the model only reads the part it needs.
• Facts a model can read. Knowledge graph statements come with labels instead of bare Q ids, best-rank values only, and dates at their real precision.
• Lists and rankings without SPARQL. Two query builders take a class, up to three filters and either a date range or a value to rank by. The query is built from ids only and refuses unfiltered lists of huge classes such as all humans.
• Exact rankings, honest limits. Values are compared in SI units where the knowledge graph normalises them, so a river given in miles and one given in kilometres rank in the right order. A query stops after 30 seconds: a very large class with a broad filter can hit that limit, and the answer then says what to narrow instead of returning a ranking that only looks right.
• Twenty identifier schemes. IMDb, DOI, ORCID, VIAF, GND, ISO codes, IATA and ICAO airports, LEI, MusicBrainz, GeoNames and more resolve to the matching item and its article.
• Attribution done for you. Each text answer names the article, links it and its history page, states the CC BY-SA 4.0 licence and says when the text was shortened.
• ChatGPT deep research gets the search and fetch pair it expects, over English Wikipedia, with up to 30,000 characters per fetched article.
• Special-category data about people (religion, health, party membership and the other categories listed under Limits) is removed before anything is cached.

Pricing (Pay-per-Event)

Event	Charged when	Price
apify-actor-start	run start (platform default)	see the Store page
tool-read (primary event)	each successful read tool call	see the Store page
tool-action	each successful call of a detailed or computed tool	see the Store page

Never charged: initialize, tools/list, pings, readiness probes, and any call that ends in an error (invalid input, not found, upstream failure, timeout, spending limit).

Platform usage comes on top of these events. This server runs in Apify Standby mode, and you also pay the platform usage of your own Standby run, billed to your Apify account like any other run: the memory set for the Standby run, for as long as the server is up, idle time included, until the run stops at its idle timeout. The Apify pay-per-event documentation says so: "For runs in Standby mode, users always pay the platform usage costs in addition to the event charges."

Limits, caveats and attribution

• Attribution. Every answer carries an attribution block. Wikipedia text is licensed under CC BY-SA 4.0: the block names the article, links to it and to its history page (the list of authors), and says when the text was converted to plain text or shortened. Wikidata is released under CC0 1.0, and the answers still name it. Definitions name the Wiktionary entry and the licence it carries, and each pronunciation file comes with its own licence and Wikimedia Commons page. Keep these lines when you reuse the content.
• No affiliation. Wikipedia, Wikidata and Wiktionary are trademarks of the Wikimedia Foundation. This server is not endorsed by or affiliated with the Wikimedia Foundation. It calls the public Wikimedia APIs, names them as the provider in every answer, and cannot guarantee their availability.
• Rate limits. Wikimedia allows 200 requests per minute to an identified client across all its APIs, and asks for one Action API request at a time. The server keeps to half of that quota per instance, runs Action API calls one at a time and at most three Wikimedia calls at once, and caches answers (search for an hour, articles and Wikidata for a day, definitions for 30 days). Heavy parallel use will still meet UPSTREAM_RATE_LIMITED.
• Wikidata queries. wiki_query_entities and wiki_rank_entities build their query from ids, years and a language code only. They run on the main Wikidata graph (scholarly articles live on a separate service and are not included), stop after 30 seconds, and refuse unfiltered listings of huge classes such as humans or taxa. Rankings use each item's best-rank value. Quantities are compared in SI units when Wikidata normalises them (lengths, areas, masses), otherwise as stated, so amounts in different currencies are not comparable.
• Personal data. For people, Wikidata statements about religion, religious order, ethnic group, sexual orientation, political party or ideology, health, criminal convictions and detention are removed before anything is cached, and these properties cannot be used as query filters or rankings.
• Scope. Articles only (no user, talk or project pages). Plain text comes from the Wikipedia TextExtracts service, which drops tables, images and references. Definitions cover English words only. The Free Dictionary API is a free, donation-funded service whose availability varies, and only its Wiktionary-sourced entries are used.
• Not included. The Wikipedia "on this day" and featured-content feeds, because the Wikifeeds API is scheduled for deprecation from July 2026, and pageview statistics.

Data terms

By using this server you agree to reuse what it returns in line with the licence of each source, and to pass the same obligations on to anyone you share it with.

• Wikipedia and Wiktionary text: CC BY-SA 4.0 (entries relayed by the Free Dictionary API carry the licence name it gives, CC BY-SA 3.0). Credit the page with its URL or its history page, state the licence with a link, say when you changed the text, and share adapted text under the same licence. The Wikimedia Terms of Use describe these requirements.
• Wikidata: CC0 1.0, no condition.
• Pronunciation audio: the licence given next to each file (for example CC BY 3.0 US or CC BY-SA 3.0).
• Trademarks: do not use the Wikimedia marks or logos in a way that suggests endorsement (Wikimedia trademark policy).
• Personal data: articles and items about living people are personal data under laws such as the GDPR, whatever their licence. Your own use of them has to respect those laws.

The fee for this server pays for the service (hosting, caching, curation, attribution), not for a licence on the content.

Use cases

• Check a fact before it goes out. A writing assistant looks up a founding date or a population in the knowledge graph and quotes the article it came from.
• Read in another language. wiki_get_article_languages lists every edition of an article with its title and URL. Albert Einstein had 235 of them in the platform test, so an agent can compare the French and Japanese versions.
• Clean up a catalogue. Map IMDb ids, DOIs, ORCIDs or ISO codes to knowledge graph items and articles, one wiki_resolve_external_id call per value. IMDb id tt0111161 resolves to item Q172241.
• Describe a neighbourhood. A travel or real-estate assistant lists the landmarks and streets that have an article near a point, sorted by distance.
• Build lists for content. The films of a director, or the most populous countries of the world: one query builder call each, with a link per item.
• Teach vocabulary. A language app asks wiki_define_word for "serendipity" and gets the pronunciation, parts of speech, definitions and synonyms, with the Wiktionary licence attached.

Frequently asked questions

How much does a call cost?

You pay per successful call, and you pay the platform usage of your own run. Each successful tool call is one tool-read event at $0.005. The two query builders, wiki_query_entities and wiki_rank_entities, are charged as tool-action at $0.01 instead, because each one runs a knowledge graph query of up to 30 seconds. Apify also charges its apify-actor-start event each time your Standby run starts.

In events alone, 100 ordinary calls come to $0.50, and 100 calls that include 20 rankings come to $0.60. These are the FREE plan prices: Apify plan discounts lower them, and the Pricing tab shows the price on your plan. On top of the events, Apify bills the platform usage of your own Standby run at your plan's rate, as it does for every pay-per-event Actor in Standby mode: the memory set for the run (256 MB by default), for as long as the server is up, idle time included, until the run stops after about 5 minutes without a call.

Errors are never charged, and neither are initialize, tools/list, pings or readiness probes. To cap the event charges of one run, set maxTotalChargeUsd. The price table by Apify plan is on the Pricing tab.

Do I need a Wikipedia account or an API key?

No. Every source is public and keyless. You only need your Apify API token to reach the server, as shown in the Connect section.

What does the server not do?

It never edits a page. It does not run raw SPARQL, does not cover the scholarly articles graph, and gives no pageviews, trending pages or "on this day" feed. Definitions are English only. A ranking of a very large class with a broad filter, such as the mountains of a whole country, can hit the 30-second limit: the error says what to narrow, and the same query is refused for 15 minutes. It also cannot guarantee that the Wikimedia services are up.

Can I reuse the text in my own product?

Yes, under the source licences. Wikipedia and Wiktionary text is CC BY-SA 4.0: keep the attribution block that comes with each answer and share adapted text under the same licence. Wikidata facts are CC0. What you pay for is the service, not a licence on the content.

How does it handle data about living people?

Articles and knowledge graph items about people are personal data under laws such as the GDPR. Before anything is cached, the server removes religion, religious order, ethnic group, sexual orientation, political party or ideology, health, criminal convictions and detention from knowledge graph answers about people, and those properties cannot be used as filters. Your own use of the answers has to respect data-protection law.

Why is my first call slower than the next ones?

When no Standby run of yours is up, Apify starts one, which takes a few seconds. After that, answers come from the source or from the cache: search results are kept for an hour, articles and facts for a day, definitions for 30 days.

Is the server affiliated with Wikipedia or the Wikimedia Foundation?

No. Wikipedia, Wikidata and Wiktionary are trademarks of the Wikimedia Foundation, which does not endorse this server. It calls the public Wikimedia APIs and names them as the provider in every answer.

Your feedback

Found a wrong answer, a missing identifier scheme or a bug? Open an issue on the Issues tab.

More MCP servers, scrapers and data studies from the same team: mr-bridge.com.

Changelog

• 0.1.11: Output, dataset and key-value store schemas, as Apify asks for publication: the Output tab of a run started from the Console shows the connection details and one row per tool.
• 0.1.10: the Input tab shows the endpoint and how to connect an AI client, and a run started from the Console outputs the connection details and the tool list
• 0.1.9: pricing events tool-read and tool-action
• 0.1.8: the wiki_rank_entities description gives the most populous countries as a ranking example instead of the highest mountains, a query that can exceed the 30 s limit, and the Pricing section names the platform usage of your own Standby run
• 0.1.7: cache keys follow the shape of the cached data, so Wikidata entities and dictionary entries stored by earlier builds are read again once, and each upstream retry or timeout logs one line
• 0.1.6: the Wikimedia APIs identify this server by its application URL, and a contact e-mail sent by the caller never reaches them
• 0.1.5: rankings stay exact, and the wiki_rank_entities description says that a very large class with a broad filter can exceed the 30 s query limit
• 0.1.4: wiki_get_entity lists the aliases Wikidata keeps under the multilingual mul code, wiki_get_article_languages returns the edition code of each link (als rather than gsw), rankings run as a named subquery, and TIMEOUT hints list only what the call can still change
• 0.1.1: first build on Apify (private, unpublished)
• 0.1.0: initial build (private, unpublished)