Web, JSON and MCP
Different readers can consume the same current public content.
The normal browser view uses a lightweight site shell and skeleton, then fetches the article separately. Internal navigation updates the main region while keeping the title bar. Related template searches load separately when their panel approaches the viewport. See Navigation and loading.
Plain HTTP readers retain full server-rendered HTML. A browser can explicitly request it with ?view=server, and the no-JavaScript view offers a fallback link. Server fallback can wait for optional template search work; it is not the same loading path as the progressive browser view.
The matching JSON route for /guide is /api/pages/guide. It returns public content and route rather than layout HTML. MCP's get_page reads the public page by route. Neither JSON nor MCP page reads execute template searches. Templates may come from the warm template cache; page selection remains live.
MCP also exposes list_pages, search_pages and get_related_pages. Anonymous visitors use these read-only operations; a public page read never becomes unrestricted access to drafts or raw storage.
The website has built-in Browse, Search and a dynamic /sitemap.xml. The sitemap is bounded discovery, not a backup or exhaustive snapshot at every corpus size. Search-result hydration and media handling retain their own public eligibility rules.
Link people to normal pages, and give agents canonical routes and public MCP tools. The connection guides help readers use the same content through their own assistants.