Docs for LLMs
Every page here is also published in a form an AI assistant reads directly, so you can hand the documentation to ChatGPT, Claude, Cursor or an agent without copying anything by hand.
| File | What it is | Use it when |
|---|---|---|
llms-full.txt | every page of the documentation in one file | you paste context into a chat, or attach it as a file |
llms.txt | an index: every page with a one-line summary and a link | a tool fetches pages itself and should read only what it needs |
<page>.md | any single page as plain Markdown | you need one page, without the site around it |
All of them are written every time the documentation is published, from the same pages you are reading. They never describe a version of the docs that no longer exists.
Any page as Markdown
Add .md to the address of a page:
https://docs.arclip.design/creators-engine/ai/mcp ← the page
https://docs.arclip.design/creators-engine/ai/mcp.md ← the same page as Markdown
Translated pages work the same way: https://docs.arclip.design/ru/creators-engine/ai/mcp.md.
llms.txt and llms-full.txt cover the English site.
Pointing your tool at the docs
A chat — ChatGPT, Claude, Gemini. Attach llms-full.txt, or, if the chat can open links,
give it the index and your question:
Read https://docs.arclip.design/llms.txt and answer from the AR Clip docs: <your question>
Cursor. Add https://docs.arclip.design/llms-full.txt as a doc in Cursor's settings, under
indexing and docs, then mention it with @ in a chat.
Claude Code, Codex and other agents. They fetch pages themselves; tell them where to start:
"read https://docs.arclip.design/llms.txt first". A folder set up with
arclip link already tells the agent this in its CLAUDE.md.
Brief for assistants is the first entry in
llms.txt: the mental model, the naming rules and the mistakes that produce confidently wrong
answers. It is written to be handed to a model as it is.
When reading is not enough
Documentation tells a model how the platform works. To let it work on your project — build the scene, wire events, check its own result — connect it over MCP: its tools answer from the live engine, so it can look up exactly which components, events and nodes exist.