# Set up in your agent

Set up AI SEO Tracker in your agent, then run one real job.

You need:

- An AI SEO Tracker account with at least one project that has **finished a scan**
- An MCP client (Cursor, Claude Code, Claude, Grok, ChatGPT, and similar)

There are no static API keys. Auth is browser OAuth from the MCP client.

## Fastest path: fetch Setup

Agents can load MCP steps, tool order, and rules from one file:

```text
Fetch https://aiseotracker.com/SKILL.md and follow it to connect AI SEO Tracker MCP and summarize my AI visibility for [your-domain.com].
```

Full reference: [Setup](/docs/setup-skill).

## 1. Create a project

Sign in, create a project, wait until prompts and sources exist. Empty tool results with `dataStatus: "no_data"` mean the scan has not landed. That is expected, not a broken connection.

Open **Settings → Agent** at `/p/{projectSlug}/settings/agent`. Copy the config shown for your client.

## 2. Add the MCP server

Production URL:

```text
https://aiseotracker.com/api/agent/mcp
```

Prefer the URL from Agent settings so local and custom domains stay correct. Do not add an `Authorization` header.

### Cursor

Save as `.cursor/mcp.json` in the repo (or Settings → MCP):

```json
{
  "mcpServers": {
    "aiseotracker": {
      "url": "https://aiseotracker.com/api/agent/mcp"
    }
  }
}
```

Reload the window. Settings → MCP → **Connect** next to `aiseotracker`. Sign in and Allow.

### Claude Code

```bash
claude mcp add --transport http aiseotracker https://aiseotracker.com/api/agent/mcp
```

Restart if the server does not appear. Approve OAuth in the browser.

### Claude, Grok, ChatGPT

Add a custom MCP / connector with **only** the URL. Connect, then sign in and Allow.

The first request without a token is `401` with `agent_auth_required`. That is how the client discovers OAuth. Details: [Auth](/docs/auth).

## 3. Test it

A test run reads live project data. It does not send email or start a scan.

Paste this in the same chat (Agent settings copies it with your slug filled in):

> Call aiseo_list_projects. Use the first project. List missing prompts and the top citation domains. Summarize what to do this week. Do not invent ranks.

Review:

- Did a login window open, then succeed?
- Did `aiseo_list_projects` return slugs you recognize?
- If prompts/sources are empty, is `dataStatus` `no_data`? Wait for a scan rather than retrying OAuth.

If the client never opens a browser, see [Troubleshooting](/docs/troubleshooting).

## 4. Pick a job

Pick one job and run it from live MCP data:

| Job | Start with |
| --- | --- |
| Report | Fetch [`/ai-seo-report/SKILL.md`](/ai-seo-report/SKILL.md) |
| Content coverage | Fetch [`/content-coverage/SKILL.md`](/content-coverage/SKILL.md) |
| Monday recap | List missing prompts and top citation domains |
| What to write | Prompts with `status=missing` |
| Who to pitch | [`aiseo_list_sources`](/docs/tools/aiseo-list-sources) |
| Why a rival wins | [`aiseo_list_competitors`](/docs/tools/aiseo-list-competitors) |
| Reddit threads AI cites | Sources with `domain=reddit` |

## Next steps

- [Setup](/docs/setup-skill), [Report](/docs/ai-seo-report), [Content coverage](/docs/content-coverage)
- [Connect MCP](/docs/mcp)
- [Tools](/docs/tools)
- Revoke at [/account/agents](/account/agents)