# Agents

What eoria gives a coding agent, and which part to set up first.

[Documentation](https://eoria.adamtrip.pt/agents/overview/)

eoria copies component source into your app, so an agent that works in your repo can read every
component it uses. What it cannot see from there is the registry, the docs and how your app is
set up. Three things fill that gap. None of them needs an account and nothing runs on a server.

| What                                                      | What it does                                                                   | Set it up when                                  |
| --------------------------------------------------------- | ------------------------------------------------------------------------------ | ----------------------------------------------- |
| [MCP server](https://eoria.adamtrip.pt/agents/mcp.md)     | Seven tools: search and read the registry, read docs, check the app, add files | Your agent supports MCP. Start here.            |
| [Skill](https://eoria.adamtrip.pt/agents/skill.md)        | Instructions that teach the agent how eoria works and what to check first      | Your agent loads skills, such as Claude Code    |
| [Markdown docs](https://eoria.adamtrip.pt/agents/docs.md) | Every page as Markdown, plus `/llms.txt` as the index                          | Nothing to set up. Point the agent at the index |

## The short version

In the root of your app:

```sh
npx @eoria/cli mcp init --client claude
npx skills add adamtrip-solutions/eoria
```

The first line writes the MCP entry for Claude Code. `--client` also takes `cursor`, `vscode` and
`codex`. The second line installs the skill. After that a prompt such as "add a settings screen
with a sign out row" has what it needs. The agent searches the registry, reads the page for the
block, copies the files and tells you which packages to install.

## One rule for every agent

The files in your app are the truth. You may have edited a component after you added it, so its
props can differ from the docs. The skill and the MCP server both tell the agent to read the
installed file before it writes code against it. If you write your own agent instructions, say
the same.