API, bulk download and MCP
Everything on this site is also a static JSON file. No key, no rate limit, no account, CORS open to any origin. Each entry carries the exact Wikipedia revision id its sentence was quoted from, so anything you build on it can be checked back to a source rather than taken on trust.
That last part is the whole point of taking this rather than scraping an encyclopedia article. An article is a moving target: quote it today and the sentence may not be there next year, which makes the citation worthless. A revision id never changes.
Endpoints
/api/year/1066.json— one year: lead, sections, entries, citations/api/date/september-7.json— one calendar day across every year/api/years.json— the index of every published year/api/dates.json— the index of all 366 dates/api/century/11th-century.json— one century: its decades, subjects and highlights/api/decade/1060s.json— one decade: its years and their leading entries/api/topic/science-and-discovery.json— one subject across every century/api/topic/science-and-discovery/17th-century.json— one subject inside one century/api/cross.json— the index of every century, decade and subject/api/index.json— the endpoint list, as JSON/api/openapi.json— OpenAPI 3.1 document
Subject timelines
/api/entity/constantinople.json— one subject: every dated entry naming it, in order, with its Wikidata id/api/entities.json— the index of every subject that has a timeline/api/search?q=plague&from=1300&to=1400— subjects matching a query and their dated entries — the one endpoint that computes
Years use astronomical numbering, the same as the URLs: /api/year/-43.json is 44 BCE. The subject and century slugs are the ones in the page URLs, so anything you can read you can also fetch. There is also a one-tag embed if you only want today’s entries on a page.
Take the whole thing
You do not have to crawl this site to have it. The entire corpus is three gzipped newline-delimited JSON files, refreshed with every build, free for any use including training and redistribution. Downloading them is cheaper for you and for us than fetching several thousand pages, and it is the version we would rather you cited.
/dump/events.ndjson.gz— every dated entry, one JSON object per line/dump/years.ndjson.gz— one record per year, with its source revision/dump/entities.ndjson.gz— one record per subject timeline/dump/datapackage.json— Frictionless data package describing all three/llms.txt— what this site is, for a machine that found it on its own
MCP server
https://slashyear.com/mcp is a Model Context Protocol server over streamable HTTP. No key, no account, no session. Point any MCP client at it and a model gets five tools — search_history, get_year, get_day, get_timeline, list_subjects — that return rows carrying the revision id each sentence came from, so the answer it writes can cite something that will still say the same thing next year.
claude mcp add --transport http slashyear https://slashyear.com/mcpExample
curl -s https://slashyear.com/api/date/september-7.json \
| jq '.groups[0].items[0]'