Connect your AI agent

Each site has a read key. Your agent uses it to read that site's analytics over MCP (Model Context Protocol) and to create campaign links. It can't change your account or billing. The key is shown once when you add a site; generate a new one any time from the site's Setup tab, which also shows these commands with your key filled in.

Claude Code

claude mcp add --scope user --transport http agentlytics https://a.measuremy.site/mcp --header "Authorization: Bearer YOUR_READ_KEY"

Restart Claude Code, or run /mcp to reconnect. claude mcp list should show agentlytics as connected. --scope user makes it available in every project.

If a tool call says the server rejected the Authorization header, the read key in your config is out of date (it changes when you rotate it on the Setup tab or re-run agent setup for the same domain). Run claude mcp remove agentlytics, add it again with the new key, then /mcp to reconnect. A running session keeps the old key until you do.

Cursor

Add to .cursor/mcp.json in your project, or ~/.cursor/mcp.json for all projects:

{
  "mcpServers": {
    "agentlytics": {
      "url": "https://a.measuremy.site/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_READ_KEY"
      }
    }
  }
}

Codex

Add to ~/.codex/config.toml:

[mcp_servers.agentlytics]
url = "https://a.measuremy.site/mcp"
bearer_token_env_var = "AGENTLYTICS_READ_KEY"

and run export AGENTLYTICS_READ_KEY=YOUR_READ_KEY in the shell that starts codex.

Share your analytics: /make-public

Agent setup through /llms.txt adds two slash commands. /make-public publishes a dashboard of your traffic, spikes and what caused them at https://measuremy.site/share/your-domain; growth actions, campaigns and setup stay private. /make-private turns it off. To add them by hand, save the text below as .claude/commands/make-public.md (Claude Code), .cursor/commands/make-public.md (Cursor) or ~/.codex/prompts/make-public.md (Codex, run as /prompts:make-public):

Make this site's agentlytics analytics public: call the agentlytics MCP tool set_public_dashboard with public=true. Then give me the public link and say in one line that it shows traffic, spikes and what caused them, while growth actions stay private.

Without the command, just ask: "make my analytics public". In Claude Code it's also available as /mcp__agentlytics__make-public.

What to ask

1. Understand. "Give me an overview of the last 30 days and explain the biggest traffic spike."
2. Ask about visitors. "How many iOS visitors from the US did we get last week?" "Where do my Firefox users come from?" "Show the last 10 signups with their country and browser." The agent calls query_visits and recent_visits.
3. Plan. "Use the plan_marketing prompt" or "Plan this week's marketing from what's working."
4. Execute. "Write the posts, make a tracked link for each, and log what we ship." The agent calls create_tracked_link and add_annotation, and can edit pages in your repo directly.
5. Measure. A few days later: "Which campaigns worked?" The agent calls get_campaign_results and suggests what to repeat or stop.

Ask about your visits

Two tools answer questions about who visits and what they do. query_visits counts the pageviews (or custom events) and unique visitors that match your filters, and can break the result down. recent_visits lists the latest individual visits. Your agent picks the arguments from your question, so you can just ask in plain words.

Examples: signups per country is event=signup, group_by=country. Where Safari visitors come from is browser=Safari, group_by=source. Daily views of the pricing page is page=/pricing, group_by=day.

Only counts and the details above come back. agentlytics stores no IP addresses or user agents, so your agent can't see them. Each visit row has a short visitor code that changes every day: it shows which rows on the same day came from one browser, nothing more.

All tools

Plus the plan_marketing prompt, which walks the agent through evidence → plan → execute → measure.