Between get_article, get_summary, summarize_article_section and summarize_article_for_query, you choose how much text enters the session — and the query-shaped summary is the one that changes behaviour, since it returns the passage about your question rather than the article's opening paragraph. Two things to set up front: the language edition is fixed when the server starts, so covering more than one means running more than one entry, and a busy setup should carry a Wikipedia access token, which is what keeps rate-limit 403s away.
A reader for Wikipedia's API with one language edition per running server. Every operation is exposed twice, under its plain name and a wikipedia_-prefixed alias, so it stays distinguishable when several servers are connected at once.
- search_wikipedia returns matching titles with snippets, with limit controlling how many
- get_article returns the full article; get_summary returns the lead summary alone
- summarize_article_for_query builds the snippet around your question rather than the article's own opening, with max_length setting its size
- summarize_article_section summarises one named section instead of the whole page
- extract_key_facts returns a list of facts, optionally narrowed to a topic within the article, with count setting how many
- get_sections, get_links and get_coordinates return the article's structure, its outbound links, and its location where it has one
- get_related_topics follows links and categories to suggest where to read next
- Language chosen at launch with --language, including variants such as zh-hans, zh-tw, sr-latn — or by country with --country US, --country Taiwan, --country Japan; --list-countries prints what is supported
- test_wikipedia_connectivity reports the base API URL, language, site information and response time in milliseconds, which is what turns a silent failure into a diagnosis
- --enable-cache caches API responses, and --access-token (or WIKIPEDIA_ACCESS_TOKEN) uses a Wikipedia personal access token to avoid rate-limit 403s
No account needed to read. It is a Python package — pipx install wikipedia-mcp puts the wikipedia-mcp command on PATH, which is what Claude Desktop needs; pip install wikipedia-mcp works too, but then the config may need the full path from which wikipedia-mcp. Default transport is stdio; --transport http or streamable-http takes --host, --port and --path, and sse remains for older clients. Exposing it on a network is where credentials appear: --auth-mode static with --auth-token, or --auth-mode jwt with --auth-jwks-uri and --auth-issuer.
One command — pipx install wikipedia-mcp