# SEO Autopilot, Community Edition

Local SEO autopilot for Cursor and Claude Code. It crawls your own site, writes an on-page checklist, and waits until you approve it. It does not publish changes, and it does not call paid rank APIs.

Бесплатная локальная версия SEO-автопилота [Track To](https://track-to.pro/seo-autopilot.html). Обход своего сайта, замечания по страницам и чеклист, который вы подтверждаете сами. На сайт ничего не выкладывается.

## What you get

- An MCP server: campaign, crawl, checklist, approve or reject, export markdown, `llms.txt` draft.
- A one-page check: `seo-autopilot audit https://your-site`.
- Data stays in `~/.seo-autopilot` (SQLite).

## What this edition does not do

The private autopilot keeps analytics connectors (Search Console, GA4, Yandex), the daily worker, and publish-after-approval. This repository does not include them. The crawler stays on the site you named, skips private networks, and stops at 40 pages. A checklist is not traffic.

## Install

```bash
git clone https://github.com/truemanD/seo-autopilot-community.git
cd seo-autopilot-community
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
```

Cursor (`~/.cursor/mcp.json`), or copy [examples/cursor-mcp.json](examples/cursor-mcp.json):

```json
{
  "mcpServers": {
    "seo-autopilot": {
      "command": "/absolute/path/to/.venv/bin/seo-autopilot",
      "env": { "SEO_AUTOPILOT_DATA_DIR": "/absolute/path/to/.seo-autopilot" }
    }
  }
}
```

Use the venv binary. A bare `seo-autopilot` works only when that environment is already on `PATH`.

## Loop

Ask the agent:

```text
Создай SEO-кампанию для https://example.com с запросами «учёт заявок» и «заявки с сайта».
Обойди до 20 страниц, собери чеклист и покажи его. Ничего не публикуй, пока я не подтвержу.
```

Tools: `seo_campaign_create`, `seo_crawl_site`, `seo_build_bundle`, `seo_bundle_pending`, `seo_bundle_approve`, `seo_bundle_reject`, `seo_bundle_export`, `seo_geo_prepare`.

`seo_bundle_export` writes a markdown file after approve. You apply the edits. `seo_geo_prepare` writes an `llms.txt` draft and a FAQ block with empty answers, so the model does not invent them.

```bash
seo-autopilot audit https://example.com
```

## Development

```bash
pip install -e ".[dev]"
pytest
```

License: MIT. Questions about the founder track, not about this crawler: [track-to.pro](https://track-to.pro/).