How it works
agentlytics turns raw visits into answers to two questions: what moved my traffic? and what should I do next? You see the answers in the dashboard, and your AI agent gets them over MCP so it can act on them.
A tiny script on your pages sends each pageview and custom event.
Every hour, traffic is compared with the same hour in past weeks.
Spikes get their causes ranked, plus a short AI summary.
Your agent plans marketing, ships it with tracked links and logs launches.
Campaign results show what worked, so the next plan repeats it.
1. Tracking
The script records the page, the referrer (where the visitor came from), UTM tags on the URL, and any custom events you send, such as signups. Each visit also gets a country, and the names of the browser and operating system (such as "Firefox" and "Windows"); the IP address and the browser's full identification string are never stored. Navigation in single-page apps is counted too. Bots and crawlers are filtered out, and each visit must come from your own domain.
2. Privacy by design
No cookies and no fingerprinting across days. A visitor is identified by a one-way hash of the site, IP address and browser that changes every day, and the IP itself is never stored. You can count daily unique visitors and same-day conversions, but nobody can be followed from one day to the next.
3. Spike detection
Every hour each site's traffic is compared with the same hour of the week over the previous four weeks (newer sites use the same hour on previous days). An hour counts as a spike when it's far outside the usual range: at least 4 robust deviations above normal, 10+ extra pageviews and 50% above baseline. Consecutive spike hours are merged into one event.
4. Finding the cause
For each spike, every traffic source, landing page and custom event is compared with the same window in the baseline weeks. Whatever rose the most is listed with its extra visits, % change and share of the spike. Anything you or your agent logged in the 24 hours before (a launch, a post, an email) is listed alongside. A Cloudflare Workers AI model then writes a two-part summary, what happened and do next, using only those measured numbers.
5. From insight to action
The Campaigns tab and the get_growth_actions tool turn the data into a ranked to-do list: double down on the channel that converts best, ride a channel that's growing, promote a page that's taking off, repeat what caused past spikes, and publish when your audience is most active. Each action comes with a suggested tracked link and how to measure it.
6. Your agent closes the loop
Connected over MCP, Claude Code, Cursor or Codex can read everything above, write the posts or page changes, create a tracked link for each, log what it ships, and come back days later to check get_campaign_results. Every cycle builds on evidence from the last one.
Plans
Free: 5,000 pageviews a month and 1 site, with every feature included. When a month's limit is reached, new visits stop being recorded until the next month or an upgrade, and existing data stays available. See pricing.