What Amami is, how it works, and what it is for

Amami is an AI-native layer for website analytics. It lets a supported AI coding assistant ask questions about the analytics data a user has authorized, instead of forcing the user to leave their coding workflow, open a dashboard, select a date range, and assemble an answer by hand.

It is built on an Amami analytics service and an MCP server that connects that data to AI clients. The goal is simple: make the evidence behind a product or growth decision available where a builder is already planning, coding, debugging, and shipping.

What Amami is for

Website analytics is often separated from product work. A developer may need to switch tools, find the right report, decide which dates to compare, and then translate the result back into a decision. That friction makes small but important questions easy to postpone:

• What changed after we shipped this page or flow?
• Which pages and traffic sources are bringing the right visitors?
• Is the setup path working for new users?
• What should we inspect before changing the product again?

Amami is designed to make those questions easier to ask in natural language, while keeping the answer connected to real analytics data. It is not a dashboard replacement, a session-replay product, or an agent that changes a product automatically. The dashboard remains useful for detailed inspection and configuration. Amami helps turn a focused question into an evidence-backed starting point for the next decision.

How it works

The workflow has a few deliberate steps:

• A website sends its analytics events to Amami's analytics service.
• The user connects a supported MCP client through the Amami setup flow. Sign-in and consent happen in the browser, not in the chat.
• The MCP server runs with the user's authorized scope and keeps its local credentials outside the conversation.
• When the user asks a question, the assistant calls the relevant analytics tool. It can inspect authorized websites, aggregate traffic statistics, trends, top pages, active visitors, or traffic sources.
• The assistant returns an answer that the user can check against the selected website, date range, and underlying dashboard data.

MCP access starts read-only. Creating websites, changing tracking, or sending events requires an explicit write opt-in. That boundary matters: the assistant can help a user understand data by default, but it does not silently alter analytics configuration or product data.

What a useful result looks like

A good result is not a polished sentence with no evidence. It should make the next check or action clearer. For example:

• “Pricing-page visits increased over the selected week, while the next-step event did not. Check the setup flow before changing acquisition spend.”
• “Most new visits came from one referral source. Compare that source with the prior period before treating the increase as sustained demand.”
• “The activation path has fewer completions after the latest release. Review the relevant page, event definition, and date range in the dashboard.”

The intended outcome is a tighter loop between shipping and learning: ask a precise question, review the evidence, make one deliberate change, then measure again. AI can reduce the cost of finding and summarizing the data. People still decide what the data means, what to change, and whether the evidence is strong enough.

Start with one decision

Begin with a question that has a real decision behind it. Name the website, date range, comparison point, and product action you want to understand. That makes the assistant's answer easier to verify and keeps analytics from becoming a list of interesting but directionless numbers.

For installation and client setup details, follow the Amami MCP installation guide. For deeper investigation, use the dashboard alongside the assistant rather than treating a single summary as the final answer.