This server wraps a Python change tracker that periodically snapshots web pages and computes structured diffs when content shifts. It exposes five MCP tools: track_url to register a page, check_changes to poll everything you're watching, get_diff to see unified diffs with section-level analysis (which headings changed), get_history to browse past snapshots, and untrack_url to stop monitoring. Under the hood it leans on markgrab for HTML extraction and snapgrab for optional screenshot comparison, storing everything in SQLite. You get line counts, changed sections, and human readable summaries like "3 lines added in Introduction." Reach for it when you need Claude to monitor documentation, competitor pages, or any URL where you care about what shifted between visits, not just that something changed.
claude mcp add --transport stdio quartzunit-diffgrab -- uvx diffgrab