# The AI-readable layer

Every published site is readable by agents as cleanly as by people — the same content, in the formats machines actually consume. All of it is scoped by the same [visibility boundary](/help-center): published docs only, always.

## Per-article markdown

Append `.md` to any article URL (or request it with an `Accept: text/markdown` header) and you get the article as clean markdown — no HTML to scrape.

```
https://{slug}.help.zerobuild.ai/quickstart.md
```

## Site-level indexes

| URL | What it serves |
| --- | --- |
| `/llms.txt` | The llms.txt standard: site title, blurb, and a link list of every published article's `.md` URL |
| `/llms-full.txt` | Every published article's full markdown, concatenated — one fetch for the whole site |
| `/sitemap.md` | Human-and-agent-readable link list |
| `/sitemap.xml` | Standard XML sitemap for crawlers (omitted when the site is set to no-index) |

## For connected agents

Agents connected through [MCP](/mcp) can read the workspace's published docs directly as a resource, and the REST API exposes the same published-docs listing. An agent answering from your docs reads exactly what your customers read — nothing more.