# Documentation for AI agents (/docs/ai-agents)



Pollaris Docs publishes the same setup information in formats that AI agents can read without parsing the visual website.

## Choose an endpoint [#choose-an-endpoint]

| Endpoint                           | Contains                                               | Use it when                              |
| ---------------------------------- | ------------------------------------------------------ | ---------------------------------------- |
| `/llms.txt`                        | A compact index with links to every documentation page | The agent can fetch more pages as needed |
| `/llms-full.txt`                   | The processed text of every documentation page         | You want one complete context file       |
| `/llms.mdx/docs/<path>/content.md` | One processed documentation page                       | The agent needs a specific guide         |

All three endpoints are public and contain documentation only. They do not expose your Pollaris account, forms, or credentials.

## Give an agent the docs [#give-an-agent-the-docs]

You can ask an agent:

```text
Read https://docs.pollaris.cc/llms.txt, then help me connect Pollaris to Cursor.
Ask before changing files. Never print or commit my API key.
```

For an offline or restricted agent, download the complete file first:

```bash
curl -fsSL https://docs.pollaris.cc/llms-full.txt -o pollaris-docs.txt
```

Attach `pollaris-docs.txt` to the conversation or add it to the agent's approved knowledge source.

## Keep credentials separate [#keep-credentials-separate]

Documentation tells an agent how Pollaris works. An MCP API key grants access to your Pollaris workspaces. Do not paste that key into a prompt, documentation file, source file, issue, or chat transcript.

When the client supports environment variables, store the key there. Grant only the scopes and workspaces the agent needs. See [MCP API keys](mcp/api-keys.md) for the complete safety checklist.

## Let an agent verify setup safely [#let-an-agent-verify-setup-safely]

Start with read-only access and ask the agent to:

1. List the Pollaris workspaces it can access.
2. List forms in one expected workspace.
3. Read one draft you choose.

Only add write access after these checks pass. Saving a draft changes data; compiling and validating a form do not.