> ## Documentation Index > Fetch the complete documentation index at: https://docs.atlaswork.ai/llms.txt > Use this file to discover all available pages before exploring further. # Building with AI agents > How agents drive Atlas, and how they read these docs. Atlas treats AI agents as first-class senders. There are two things an agent needs from this site: a **tool surface** to act through, and a **machine-readable index** of the docs to ground itself in. ## Act through the MCP server Agents run the signing lifecycle through the [MCP server](/mcp/overview) — a JSON-RPC tool set with review-first defaults and idempotent writes. Start there: Why MCP, the endpoint, and auth. The signing tool set. ## Read the docs as data These docs publish a machine-readable index so an agent can discover and ingest the right pages without scraping HTML: * [`/llms.txt`](https://docs.atlaswork.ai/llms.txt) — a structured index of every page, with titles and descriptions. * [`/llms-full.txt`](https://docs.atlaswork.ai/llms-full.txt) — the full docs content as a single plain-text document, ready to drop into a context window. Any docs page is also available as Markdown by appending `.md` to its URL — for example, `https://docs.atlaswork.ai/quickstart.md`. See the [agent-readable index](/agents/index-for-llms) for a curated entry point. ## Principles agents can rely on * **Agents prepare, humans approve.** Sends are review-first by default; a human clicks Send. Don't design flows that assume an agent can finalize a send without a human gate. * **Retries are safe.** Every write is [idempotent](/guides/idempotency); the SDK and MCP server supply idempotency keys automatically. * **The reference never drifts.** The [API reference](/api-reference/introduction) is generated from the live spec, so what an agent reads matches what the API does.