About
Wikinews MCP is a product demonstration by Miso. It puts an archive of about 20,000 Wikinews articles behind a Model Context Protocol (MCP) server, so that an AI assistant — Claude, Cursor, Codex, ChatGPT and others — can search the archive and read the articles it finds, and answer with sources instead of from memory.
What it shows
Miso builds AI search and answer systems for publishers and media companies. This site is a small, public, end-to-end example of the same approach applied to an open archive:
- A content library behind MCP. The two tools,
ask_wikinewsandget_wikinews_article, give an assistant exactly what it needs to answer a question about a news event: find the relevant reporting, then read it in full before quoting. The article tool only returns what a search just surfaced, which keeps the archive from being bulk-downloaded through the assistant. - Standard sign-in. Access is granted through OAuth 2.1 with dynamic client registration and PKCE, the flow every current MCP client implements, so there are no API keys to hand out and nothing to paste into a config file beyond the server URL.
- Discoverable by agents, not just people. The host publishes the files that agents and agent platforms look for: an
llms.txtindex, an MCP server card, an OpenAPI description of the endpoint, an Agent Skills discovery index with thewikinews-researchskill, an ARD catalog, and a listing in the official MCP Registry under Miso's own domain namespace. - The skill, not only the tool. The
wikinews-researchskill, shipped with the Claude Code and Antigravity plugins and downloadable for other clients, teaches an assistant how to use the archive well: search first, read the full article before quoting figures, always cite, and say so when the archive has nothing.
The archive
The content is Wikinews, the collaboratively written news site of the Wikimedia Foundation, used under the Creative Commons Attribution 2.5 licence. The archive here runs from 2004 to the present and is densest for 2005–2010, with thinner coverage in later years. It is a fixed snapshot refreshed from time to time, not a live feed, so it is not the right tool for breaking news.
Miso is not affiliated with Wikinews or the Wikimedia Foundation. This site is a demonstration; it is not an official Wikinews service.
Who makes it
Miso Technologies, Inc. ("Miso") is based in San Francisco and makes AI search, recommendation and answer products for publishers. The company site is miso.ai. You can reach the team through the contact page.
Where to go next
- Install guide — connect the server to your assistant in a minute
- Developers — the endpoint, authentication, tools and examples
- Privacy — what this demo records when you use it