# Connect your agent

This site is built to be read by humans **and** by agents. If you use Claude Code, Cursor, Codex or any other agent to build with the Fayz SDK, point the tool at the artifacts below — that way the agent answers with the SDK's real structure instead of guessing.

{% callout type="tip" %}
This page is about connecting an agent **to the Dev Center**. If you want to understand how your own app becomes a surface an agent reads and operates, see [Your app as a data layer](/en/docs/ai/your-app-as-a-data-layer).
{% /callout %}

## Every generated app ships with an `AGENTS.md`

When you run `fayz create`, the scaffold includes an **`AGENTS.md`** at the project root: a **vendor-neutral** agent guide describing the app structure, the customization checklist and the Supabase walkthrough. Any agent that respects the `AGENTS.md` convention reads that file and immediately knows that "customizing = editing the manifest, not writing new code", where the files live and how to connect a real database.

The scaffold also generates a `CLAUDE.md` — but it's just a **pointer** to `AGENTS.md`, so Claude Code doesn't duplicate the guide. There's only one source of truth.

{% callout type="tip" %}
The generated `AGENTS.md` points the agent back to this Dev Center — to `llms.txt` (the map) and to the site (the human reference). So an agent that opens your app already knows where to find the rest.
{% /callout %}

## The entry point: `/llms.txt`

Start with [`/llms.txt`](/llms.txt). It lists the Dev Center's entire navigation tree in plain text, with a summary of each area — the map an agent reads first to learn what exists.

```bash
curl https://developers.fayz.ai/llms.txt
```

If the agent needs **everything at once**, there's [`/llms-full.txt`](/llms-full.txt): the raw Markdown of every page, concatenated in navigation order. It's heavy — prefer `llms.txt` as the index and pull specific pages on demand.

## Every page in Markdown

Every documentation page has a **raw Markdown** version, served at the same path with a `.md` suffix. At the bottom of each page there's a **"View this page in Markdown"** link — that's exactly this file. Point the agent at the `.md` of the relevant page when you want to give it specific context without the HTML in the way.

For example, this page in Markdown lives at `/en/docs/ai/connect-your-agent.md`.

## Example prompts

Once you've given the agent `llms.txt`, short prompts work well:

```
Read https://developers.fayz.ai/llms.txt. Then add the crm plugin
to my app.manifest.json and run fayz doctor to validate.
```

```
My npm run dev opens a blank screen. Read the Dev Center troubleshooting
page in Markdown and tell me whether that's expected.
```

```
Connect this app to a real Supabase following AGENTS.md: fill in
.env.local and run fayz db apply --dry-run before applying for real.
```

```
Read /en/docs/ai/your-app-as-a-data-layer.md. My own plugin
has an orders registry — propose the aiTools I should declare
in the manifest, with mode and permission.
```

## Best practices

- Give the agent `llms.txt` first, then specific pages on demand.
- Prefer the Markdown version of each page over the rendered HTML.
- Keep the app's `AGENTS.md` in context — it carries the project-specific conventions.
- Ask the agent to cite the slug of the page it used, so you can check it.
- Once connected, point it at the ready-made procedures in [Skills](/en/docs/ai/skills) — official Fayz skills and community skills.

## MCP: coming soon

A dedicated Fayz **MCP server** — so the agent can query the catalog and the docs without pasting URLs — is **coming soon** (in design). For now, `llms.txt` plus the `.md` pages are the official path.

---

Next: [Your app as a data layer](/en/docs/ai/your-app-as-a-data-layer) · [Connectors and channels](/en/docs/ai/connectors-and-channels).