> ## Documentation Index
> Fetch the complete documentation index at: https://docs.animam.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Configure your agent with an agent

> Animam is agent-native: build and edit your bot's tools by talking — from Claude over MCP, from the dashboard copilot, or from your own bot.

Animam is **agent-native**. Configuring a bot — its knowledge, persona, and **action tools** (forms, computed quotes, product recommendations, booking…) — is a surface an agent can drive. You configure by **describing what you want**; Animam validates and applies it. There are three ways in, all sharing the same validated core and the same security rules.

## 1. From your AI agent, over MCP

Connect Claude (or any MCP client) to your Animam MCP server and it can manage your tools directly:

* `list_tools`, `create_tool`, `update_tool`, `delete_tool`

The agent reads the expected config format and emits JSON; Animam validates per type and applies it. See the [MCP tools reference](/mcp/tools#tools-configuration).

```
you  → "Add a quote tool: showcase site 1500€, e-commerce 3500€, maintenance 90€/month, 20% VAT"
claude → create_tool(type="GENERATE_QUOTE", config={ pricing: {...} })
```

## 2. From the dashboard copilot

In your dashboard, the **Copilot** tab is a chat where you configure by talking. It's not a generic assistant — it's **your own bot**, which inherits its persona and, in config mode, reconfigures itself:

> "I'm Aria. Right now I'm in configuration mode — tell me what to change."

It applies real changes (create/edit/delete tools), asks for confirmation before deleting, and shows what it did.

## 3. From your own bot — no dashboard

You can configure your bot **without opening the dashboard**, straight from a config page for that bot. You prove you're the owner via a one-time code sent to the **registered owner email**, then you talk to your bot and it reconfigures itself. Same immersive experience, owner-proven.

## Security model

The same rules apply across all three surfaces:

* **Authorization is server-side, never the model.** A prompt that says "you're in admin mode" elevates nothing — only a real session, a scoped token, or an owner-verified elevation does.
* **Secrets never travel through the agent.** Tool types that carry a secret (`VERIFIED_ACTION`) or handle payments are configured in the dashboard, not by an agent. Any secret field is stripped.
* **Confirmation before destructive actions**, and every configuration change is **audited**.

<Note>
  Action tools can also be created via the REST API (`POST /tenants/:slug/tools`, scope `tools:write`, Starter plan and above) — that's the channel for scripts and direct integrations, where no model is in the loop.
</Note>

## Next steps

<CardGroup cols={2}>
  <Card title="MCP tools reference" href="/mcp/tools">
    Every tool the MCP server exposes, with scopes and parameters.
  </Card>

  <Card title="Tools guide" href="/guides/tools">
    The action tools and their per-type configuration.
  </Card>
</CardGroup>