Windrose.

Why it exists

AI agents increasingly build and maintain websites, but traditional analytics tools still make them work through dashboards or verbose APIs built for humans.

What I built

• A lightweight cookieless tracker with privacy-minimized visitor and country data
• An MCP server with compact tools for summaries, comparisons, anomalies, and AI traffic
• Server-side crawler capture for non-JavaScript agents such as GPTBot and ClaudeBot
• A proactive path from traffic evidence to a suggested, human-reviewed code change

Where it stands

Windrose is live at windrose.sh and tracks both human visits and AI crawler activity. Its product claims are backed by executable tests rather than marketing copy alone.

What I learned

Agent-native software needs more than an MCP wrapper. The useful layer is compact evidence, clear uncertainty, and actions an agent can safely take next.