# RankCanon

MCP server that gives AI agents **fresh, cited SEO knowledge from primary sources** instead of stale training data and SEO-blog myths.

## Sources (v0.1)

- **Google Search Central** documentation (`developers.google.com/search/docs`)
- **Google Search Status Dashboard** — core, spam and other ranking updates with dates

Planned: schema.org / structured data requirements, web.dev (Core Web Vitals), Bing, Yandex, doc change diffs, semantic search, project-specific knowledge.

## Tools

| Tool | Purpose |
|---|---|
| `search_docs(query, limit, source)` | Full-text search (SQLite FTS5) with snippet, URL and last-changed date |
| `get_doc(url)` | Full stored text of a page |
| `algorithm_updates(date_from, date_to)` | Google updates overlapping a date range, to correlate traffic drops |
| `corpus_status()` | What is indexed and how fresh it is |

## Quick start

```bash
uv sync
uv run rankcanon-sync                 # fetch sources (use --max-pages N, --source docs|status)
uv run rankcanon                      # run MCP server over stdio
```

Database location: `~/.local/share/rankcanon/rankcanon.db` (override with `RANKCANON_DB`).

### Claude Code

```bash
claude mcp add rankcanon -- uv --directory /path/to/rankcanon run rankcanon
```

Run `rankcanon-sync` on a schedule (e.g. daily cron) to keep the corpus fresh.

## License

MIT