fuma-nama/fumadocs fumadocs@16.19.1 on GitHub

llms() adds files() and serve(path): llms.txt, llms-full.txt and <page>.md files by their public path, served from one route.
URL directories get their own llms-full.txt, listed in llms.txt, so large docs can be read section by section.
llms.txt from serve() links pages to their Markdown.
The filter option leaves pages out of full() and the files.
Components without a Markdown form keep their MDX tag name in processed Markdown, also in production builds.
The get_page MCP tool accepts URLs with a hash or query, like the URLs of search results.
Auto type tables show the full type of each prop in Markdown.

CLI prompts don't wait without a terminal: a missing choice fails with the flag and its values, confirmations take their default, and existing files are kept unless --yes is passed.
The CLI fails with a clear message where it finds no supported framework, instead of setting up Next.js.
The CLI detects Radix UI from a plain fumadocs-ui dependency.
init --i18n on a static Next.js site keeps the locale in URLs and writes no proxy.
The duplicated slugs error names both files.
The macro error of Fumadocs MDX explains how to load content outside the bundler.
New projects include an AGENTS.md that links the Fumadocs docs for AI agents.
The CLI and create-fumadocs-app print plain lines instead of spinner frames without a terminal.
create-fumadocs-app names the file to set the site name and GitHub repository, and names the package after the current folder when created in ..
The path that a collection schema function receives is documented as absolute.