> ## Documentation Index
> Fetch the complete documentation index at: https://datum-4926dda5-docs-api-reference-demo.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents Overview

> How AI agents work with Datum Cloud — Skills, MCP, llms.txt, and datumctl.

Datum is built for agents as first-class users. Rather than wrapping an API in a chat interface, Datum exposes the same primitives humans use — YAML manifests, `datumctl` commands, and CRD-based resources — in ways that agents can reason about reliably. This section covers the four surfaces agents use to work with Datum: **Skills** (procedural know-how), **Datum MCP** (live API access), **llms.txt** (doc index), and **datumctl** (the CLI agents drive directly).

## The four agent surfaces

<CardGroup cols={2}>
  <Card title="Skills" icon="book-open" href="/agents/skills">
    Self-contained instructions agents load on demand. One skill per Datum resource — install once, agents handle the rest.
  </Card>

  <Card title="Datum MCP" icon="plug" href="/datum-mcp">
    The official Model Context Protocol server. Gives agents authenticated, live access to Datum APIs.
  </Card>

  <Card title="llms.txt" icon="file-lines" href="/agents/llms-txt">
    A single index file at datum.net/docs/llms.txt that agents fetch to discover the full documentation surface.
  </Card>

  <Card title="datumctl" icon="terminal" href="/datumctl/overview">
    The CLI. Agents invoke it directly for any platform action — same surface humans use.
  </Card>
</CardGroup>

## Which one should I use?

| If your agent needs to…                                              | Use           |
| -------------------------------------------------------------------- | ------------- |
| Know *how* to do a Datum task (e.g. attach a domain to an HTTPProxy) | **Skills**    |
| Read or write live Datum resources during a task                     | **Datum MCP** |
| Discover what documentation exists                                   | **llms.txt**  |
| Execute commands directly                                            | **datumctl**  |

These surfaces layer together. A typical agent session loads one or two relevant Skills at startup, calls `datumctl` or Datum MCP to act, and falls back to `llms.txt` when it needs deeper context. See [Skills](/agents/skills) to get started.