# Overview

> Let AI agents work in Lucra, within limits you set.

Source: https://www.onlucra.com/docs/agents

AI agents can work in Lucra the way people do: read programs and results, review work, message creators and, if you allow it, move money. They use the same API, with the same access checks.

## Connect

| Option                                  | Best for                                        |
| --------------------------------------- | ----------------------------------------------- |
| [MCP server](https://www.onlucra.com/docs/mcp)            | AI assistants; you sign in with OAuth          |
| [REST API](https://www.onlucra.com/docs/api-introduction) | Agents you run in your own code, with an API key |
| [llms.txt](https://www.onlucra.com/docs/llms-txt)         | Giving a coding agent these docs                |

## Keep agents scoped

- Grant `read` first, and `write` only once you trust the agent with every write, including money. See [Access](https://www.onlucra.com/docs/access).
- Give each agent its own key or connection, so you can revoke one alone.
- Send an `Idempotency-Key` on writes, so a retry never acts twice.