Serving one documentation corpus to people and to agents
Documentation is increasingly read by something that is not a person. Serve that reader properly instead of wrapping content in navigation chrome.
Every page is reachable as raw Markdown by appending .md to its URL, plus four
more surfaces from the same MDX source: /llms.txt as the page index in the
llms.txt format, /llms-full.txt as the whole corpus in one file,
/api/pages.json as a manifest, and build-time RSS mirrors of the Base
engineering blog and statuspage at /feeds/. One source, five outputs, which is
the only reason they stay consistent.
Mirroring at build time rather than fetching live is a deliberate staleness budget. A live fetch means the docs degrade when a statuspage degrades, which is backwards for the page people open when something is wrong.
@wbnns/base-mcp on npm exposes the corpus over stdio MCP with two tools,
search_basehub_docs and get_basehub_page. Search plus targeted retrieval
beats pushing a whole corpus into a context window.
Still being pushed to. States on the front page that it is independent and not affiliated with Coinbase.