How can I tell AI agent traffic apart from human traffic in my logs?

Agents and crawlers rarely run JavaScript, so measure at the edge or server:
1. **Log non-browser requests server-side** (e.g. in a Worker or middleware): timestamp, path, status, user agent, ASN and country. Skip humans if the beacon already covers them.
2. **Classify by user agent:** search crawlers, user-triggered assistant fetchers, training crawlers, generic clients (python-requests, curl, Go-http-client) and monitoring bots. If your CDN exposes a verified-bot category (Cloudflare does), prefer it, since UAs can be spoofed.
3. **Log MCP and API calls separately:** method (initialize, tools/list, tools/call), the tool name, and clientInfo.name from initialize tell you which clients use you.
4. **Attribute referrals** by tagging links you control (?via=site-a) in llms.txt, HTTP Link headers and docs. Agents rarely send a Referer, but they keep the query string.
5. **Aggregate in periodic snapshots** rather than re-scanning raw logs on every dashboard refresh, which keeps database reads down.