---
title: Connect an agent
description: Add your brand to Claude, Claude Code, Cursor, Codex, ChatGPT, v0, Lovable or any MCP client, sign in once, and choose whether it may edit.
---

> Source: https://www.heyarchetype.com/docs/agents/connect

Every brand in Archetype is its own MCP server. Connect it to the AI tools your team builds with, and they read your tokens, components, templates, assets and voice live, check their own work against the brand, and, if you allow it, propose changes to it.

## Your brand's MCP address

Each brand has its own address:

```text title="MCP server URL"
https://www.heyarchetype.com/api/mcp/<brand-slug>
```

You don't need to build it by hand. Open **Settings → Agents**, and under **Connect an agent** the **MCP** tab shows your brand's exact address in step 1 (titled with your brand's name, for example **Add Acme to your agent**), already filled into the snippet for the client you pick.

One address per brand is what lets a person with several brands switch between them. Add each brand as its own connector, named `archetype-<brand-slug>`, and turn the one you need on in your client. A connection made for one brand is refused at another brand's address.

The bare address, `https://www.heyarchetype.com/api/mcp`, also works. It asks which brand to share when you sign in, which is why the public setup guides use it.

## Add Archetype to your client

:::steps
### Open Settings → Agents
Any member can open it. Under **Connect an agent**, stay on the **MCP** tab and pick your client.

### Add the server
Paste the URL or run the command the tab shows. The table below says where each client takes it.

### Sign in once
Your client opens a browser tab with Archetype's approval screen. Pick the brand if it asks, decide whether to allow edits, and click **Connect**.
:::

| Client | Where you add it | Full guide |
| --- | --- | --- |
| Claude (web or desktop) | **Settings → Connectors → Add custom connector**, name it after the brand, paste the URL, then click **Connect** and approve. | [/mcp/claude](/mcp/claude) |
| Claude Code | Run the command below, then type `/mcp`, select the server and choose **Authenticate**. | [/mcp/claude-code](/mcp/claude-code) |
| Cursor | Add it to `~/.cursor/mcp.json` (or `.cursor/mcp.json` in your project), then turn it on under **Settings → MCP** and click **Connect**. | [/mcp/cursor](/mcp/cursor) |
| Codex | Run `codex mcp add archetype-<brand-slug> --url <your MCP URL>`, then approve the browser tab Codex opens. | [/mcp/codex](/mcp/codex) |
| ChatGPT | On the web, turn on **Developer mode** in Settings, create a new MCP app with OAuth, name it after the brand and paste the URL. | [/mcp/chatgpt](/mcp/chatgpt) |
| v0 | Click **Add MCP** (in Settings or the prompt form), choose a custom server and paste the URL. | [/mcp/v0](/mcp/v0) |
| Lovable | Open **Connectors**, click **+**, choose **MCP server**, keep OAuth, paste the URL and click **Add & authorize**. | [/mcp/lovable](/mcp/lovable) |
| Any other client | Any MCP client that supports remote (HTTP) servers with OAuth. | [/mcp](/mcp) |

For Claude Code, from any terminal:

```bash title="Terminal"
claude mcp add archetype-<brand-slug> --scope user --transport http https://www.heyarchetype.com/api/mcp/<brand-slug>
```

For Cursor, in `~/.cursor/mcp.json`:

```json title="~/.cursor/mcp.json"
{
  "mcpServers": {
    "archetype-<brand-slug>": {
      "url": "https://www.heyarchetype.com/api/mcp/<brand-slug>"
    }
  }
}
```

## What the approval screen asks

Signing in is OAuth 2.1, the same flow every MCP client already knows. The approval screen names the client asking (**Connect Claude Code**, for example), shows the brand, and lists what the connection lets it do:

- Read your tokens, components, assets and voice
- Search everything in the brand
- Check its own work against the brand

The last line is a checkbox: **Change the brand when you ask it to**. Tick it to let the agent call `update_brand`. Leave it clear for a read-only connection.

Ticking it only grants what your role already allows. A Viewer who ticks it still gets a read-only connection, and each change is checked against your rights in the area it touches. See [Members and roles](/docs/workspace/team).

> [!NOTE]
> A connection belongs to the person who approved it. If you leave the brand, your connections stop working, whatever client they live in.

## OAuth sign-in or an API key

MCP connections always sign in as a person, through the approval screen. The MCP server does not take API keys.

API keys are for places where nobody can sign in: CI, scripts and build steps. An admin creates one in **Settings → Agents**, on the **API** tab, with **Create API key**. The key is shown once, so copy it then and store it as a secret, usually as `ARCHETYPE_TOKEN`. The [CLI](/docs/developers/cli) and the brand's HTTP endpoints accept it.

## Permission prompts

Many clients ask your permission the first time each tool runs. Archetype publishes six tools, so you see at most six of these prompts, not one per kind of brand data.

`get_brand`, `search_brand` and `check_against_brand` are marked read-only. `generate_media`, `create_campaign` and `update_brand` are marked as writes, so a client that asks before write actions asks before these three. The [MCP tools](/docs/agents/tools) page lists every argument.

By default a change from `update_brand` doesn't land straight away. It waits in **History** for someone to apply it. See [Changes from agents](/docs/agents/drafts).

## Manage your connections

**Settings → Agents** lists the clients you have connected under **My connections**, with when each was last active. Open a row's menu and choose **Revoke** to disconnect it. Your connections are yours: other members see and revoke their own.

If the brand's plan ends, connected agents stay connected but every tool answers that the brand context is paused until an owner or admin renews it in **Settings → Plan**.

## Next steps

:::cards
- [MCP tools](/docs/agents/tools): Every tool, its arguments and an example call.
- [DESIGN.md, BRAND.md and skills](/docs/agents/files): Skills that teach your agent when to call the tools, and the files it can read.
- [Changes from agents](/docs/agents/drafts): What happens when an agent edits the brand.
:::