api

Use it from anywhere

No key, no signup. Every endpoint is a plain GET with open CORS, so it works from curl, scripts, browsers and agents.

URL prefix

Put llms.ink/ in front of any link. The scheme is optional; query strings are kept.

GET https://llms.ink/<link>        # llms.txt
GET https://llms.ink/full/<link>   # llms-full.txt

$ curl llms.ink/developers.cloudflare.com/kv/
$ curl llms.ink/full/github.com/honojs/hono > hono-llms-full.txt
$ curl llms.ink/https://arxiv.org/pdf/1706.03762

Responses are text/plain; charset=utf-8. Browsers (requests that accept text/html) are redirected to the web app instead. Headers tell you what came back:

Header	Values
`x-llms-kind`	`site` · `section` · `page` · `repo` · `document`
`x-llms-source`	`native` (published by the site) · `generated` (made by llms.ink)
`x-llms-tokens`	Approximate token count (characters ÷ 4)

Errors come back as llms.ink: <message> with a 4xx/5xx status.

JSON API

GET /api/llms?url=<link>&mode=llms|full&fresh=1

Param	Meaning
`url`	Required. Any http(s) link; `https://` may be left off.
`mode`	`llms` (default) or `full`.
`fresh`	`1` skips the 24h cache and rebuilds.

{
  "url": "https://developers.cloudflare.com/kv/",
  "kind": "section",
  "source": "generated",
  "mode": "llms",
  "title": "Cloudflare Workers KV",
  "text": "# Cloudflare Workers KV\n\n> Workers KV is …",
  "tokens": 896,
  "pages": 25,
  "model": "anthropic/claude-haiku-5.5",
  "cached": false,
  "generatedAt": "2026-11-09T18:03:51.417Z"
}

pages is the number of links in llms.txt, or the number of pages inlined in llms-full.txt. model is null for native files and when links were mapped without AI descriptions. Errors return { "error": "…" } with a 4xx/5xx status.

For agents

Any agent with a fetch tool can use llms.ink — no MCP server or plugin needed. Drop this into a system prompt or rules file:

When you need documentation for a library, product or website, fetch
https://llms.ink/<its docs URL> for an index (llms.txt), then fetch
the pages you need — or https://llms.ink/full/<URL> for everything at once.

Limits

60 new links per IP per day; cached links don’t count. 10 requests per minute.
Results are cached for 24 hours per link and output mode.
llms-full.txt inlines up to 20 pages (60k characters each).
Pages that need a login, block bots, or render only with JavaScript can come back thin.