llms.txt
llms.txt is a convention for a site index aimed at language models. Movk Nuxt Docs integrates nuxt-llms and prerenders two files at build time:
| File | Content | Use it to |
|---|---|---|
/llms.txt | Site title, summary and page links pointing at each page's Markdown | Let a model see what the docs cover first |
/llms-full.txt | The full Markdown of every page | Hand the whole site to a tool with a long context window |
Defaults
Without any configuration, the layer fills these in:
| Option | Source |
|---|---|
llms.domain | NUXT_PUBLIC_SITE_URL, NUXT_SITE_URL, or the URL your host provides (Vercel, Netlify, Cloudflare Pages, GitLab Pages); https://example.com when none is set |
llms.title / llms.full.title | site.name, then name in package.json |
llms.description / llms.full.description | description in package.json |
domain prefixes every link in the files, so make sure it is the production URL before a production build.
Configuration
Override them under llms in nuxt.config.ts:
export default defineNuxtConfig({
llms: {
domain: 'https://docs.example.com',
title: 'My Library',
description: 'Official documentation for My Library: installation, configuration and API.',
notes: [
'Append .md to any page URL to get its Markdown'
],
full: {
title: 'My Library full documentation',
description: 'The full Markdown of every documentation page.'
}
}
})
notesrenders as a## Noteslist at the end of the file/llms-full.txtis generated only whenfullis set, which the layer does by default
Page sections
Page links are read from the content collections by nuxt-agent-discovery and rewritten to their /raw/**.md URLs, the same content a reader gets by appending .md. To control the grouping, filter by collection and path with llms.sections:
export default defineNuxtConfig({
llms: {
sections: [
{
title: 'Getting Started',
contentCollection: 'docs',
contentFilters: [
{ field: 'path', operator: 'LIKE', value: '/docs/getting-started%' }
]
},
{
title: 'Guide',
contentCollection: 'docs',
contentFilters: [
{ field: 'path', operator: 'LIKE', value: '/docs/guide/%' }
]
}
]
}
})
A section carrying its own links is kept as is, and its links to pages on the site are rewritten to their Markdown URLs too.
Extra details
agentDiscovery.llms.details inserts paragraphs between the summary and the first section, a good place to tell models how to use the files. It requires llms.description, and the paragraphs must not contain headings:
export default defineNuxtConfig({
agentDiscovery: {
llms: {
details: [
'Append `.md` to any page URL to get its Markdown.',
'Read `llms-full.txt` when you need the whole site.'
]
}
}
})
Multiple languages
nuxt-llms generates a single /llms.txt per site; its title and summary are not localized. On a multilingual site, write llms.description in the language you want models to read and mention the other locales' prefixes (such as /en/docs) in details.
Relation to Markdown pages
llms.txt is only the entry point. Every page has its own Markdown representation, served by agent discovery:
# Append .md to the page URL
curl https://docs.example.com/en/docs/getting-started.md
# Or request the raw path directly
curl https://docs.example.com/raw/en/docs/getting-started.md
MDC components are converted to plain Markdown first, so models get text they can read directly.
Use it in AI tools
- Cursor: type
@Docsin the chat and addhttps://docs.example.com/llms.txtas a docs source (type the@by hand; pasting it does not trigger the menu) - Windsurf: reference the
llms.txtURL with@docs, or put it in a workspace rule - ChatGPT, Claude and others: include the URL in your question, e.g. "answer using https://docs.example.com/llms.txt"
Start with /llms.txt and let the model fetch pages as needed; use /llms-full.txt only when the tool supports a very long context and must read everything at once. For ongoing lookups, connecting the MCP Server works better.
Agent Skills
Publish the Agent Skills in your project's skills/ directory under /.well-known/skills/ together with the docs site, so readers install them into Claude Code, Cursor and other assistants with one command. Covers layout, naming, publishing and the two built-in skills.
Agent Discovery
The docs site serves AI agents a Markdown version of every page, a Markdown sitemap and a set of /.well-known discovery documents, so agents can find and read content without parsing HTML. This page lists every endpoint, the negotiation rules and what the layer adds.