AI Workflows

How to connect your analytics to ChatGPT and Claude with MCP

You can give ChatGPT, Claude, or Cursor live read access to your website analytics using the Model Context Protocol (MCP). Once connected, your AI agent can read your real traffic, drop-off, and revenue data and act on it in your codebase in the same session. Here is exactly how to set it up with AnalyzeUser.

What is MCP and why connect analytics to it?

MCP (the Model Context Protocol) is an open standard that lets AI tools call external services in a structured, secure way. Connecting your analytics to it means your AI agent stops guessing and starts reading the real numbers from your product.

Without MCP, if you ask an AI coding assistant to improve a page, it has no idea how that page actually performs. It cannot see that your pricing page has a 70 percent drop-off, or that most of your signups come from a single blog post. It is working blind.

With an analytics MCP server connected, the same agent can pull live data first, then make a change informed by what is actually happening. AnalyzeUser runs a remote MCP server for exactly this. It is read-only and scoped to your own sites, so your AI tools get insight without any risk of changing your data.

What can an AI agent do with your analytics?

Once connected, your AI agent can call a set of read-only tools that return the same numbers you see on the AnalyzeUser dashboard. The current tools are:

• list_sites - list the sites the token can read
• get_growth_report - a one-shot overview of how a site is doing
• get_stats - core metrics like visitors, pageviews, and bounce
• get_top_pages - your most visited pages
• get_traffic_sources - where visitors come from, including a dedicated AI channel
• get_revenue - revenue, including revenue by source
• get_exit_pages - where people leave
• get_top_locations - geography of your visitors
• get_devices - device, browser, and OS breakdown

Here is the payoff in one workflow: ask Claude which page has the worst drop-off, let it call get_exit_pages, and then have it fix the copy on that page in the same session. The agent that reads the problem is the agent that ships the fix.

How to connect your analytics to ChatGPT and Claude

Connecting takes a few minutes and three steps. You create a token, paste a config into your AI tool, and let the agent connect.

1. Create a personal access token on the /ai-tools page. The token starts with au_mcp_ and is shown only once, so copy it right away.
2. Pick your AI tool (Cursor, Claude Code, ChatGPT, or Windsurf) and copy the MCP config the page generates for it. Paste that config into your tool's MCP settings.
3. The agent connects to the AnalyzeUser MCP server. Ask it something like "what are my top traffic sources this week" and it will call the read-only tools and answer from your live data.

That is it. The setup wizard on /ai-tools even detects when your tool actually connects, so you can confirm the link works before you rely on it.

Is it safe to give an AI agent your analytics?

Yes. The AnalyzeUser MCP server is read-only by design, so an agent can read your data but cannot write, edit, or delete anything.

• Read-only - the tools only fetch aggregations. There is no write path.
• Scoped to your own sites - ownership is re-checked on every call, so a token can never read another account's data.
• Revocable - delete the token on the /ai-tools page and access stops immediately.
• No PII - AnalyzeUser is privacy-friendly and collects no personal data, so you are sharing aggregate analytics, not customer identities.

AnalyzeUser vs other analytics MCPs

Most analytics MCP servers are thin. They either help an agent install a tracking snippet, or they let you look up a single visitor by id. Useful, but not insight.

AnalyzeUser's MCP server is insight-first. It exposes the same SQL aggregations the dashboard runs, so an agent can see your top pages, drop-off, traffic sources, and revenue by source, then act on them in your codebase. Because the MCP and the dashboard read through the same queries, the numbers an agent reports always match what you see in the UI. There is no second source of truth to reconcile.

Try AnalyzeUser free for 14 days

No credit card. No setup fee. Paste one snippet, then connect the MCP server so ChatGPT, Claude, or Cursor can read your live analytics while it codes.

Frequently asked questions