Blog
Analytics MCP server for Claude and ChatGPT
Millions of people already work inside Claude, ChatGPT, Copilot, or Gemini. MetriXs ships a first-party MCP server, so you can ask those assistants about your traffic and get your real numbers instead of a guess.
MetriXs editorial team · Published
What MCP is, in one paragraph
MCP (Model Context Protocol) is the open standard that lets AI assistants use external tools. You connect an assistant to a server once, and from then on the assistant can call that server's tools whenever a question needs them. Ask "how many visitors did my site have last week?" and the assistant calls the analytics tools, reads the real data, and answers in plain language, with the numbers from your dashboard instead of something it made up.
MetriXs runs its own MCP server at mcp.metrixs.eu. It is first-party: no third-party proxy in the middle, no plugin to trust, the same company that stores your data is the one answering questions about it.
The questions it answers
Once connected, plain-language questions work against your live data:
- "How many visitors did I get this week compared to last week?"
- "Which pages brought in the most visitors this month, and which sources sent them?"
- "How many people are on the site right now?"
- "What did my custom event 'signup' do this month, and which plans did people pick?"
- "How is my checkout funnel doing, and what is my best-selling product?" (commerce sites)
- "How many visitors came from ChatGPT or Perplexity this month?" (AI referrals)
Every answer comes from the same queries your dashboard uses. The assistant never writes SQL and never sees raw rows, only the aggregates the tools return.
Two ways to connect
API key for code assistants: Claude Code, OpenAI Codex, Gemini CLI, VS Code with GitHub Copilot, Cursor, and JetBrains. You create a site-scoped API key in Settings, paste it into your client's MCP config with the endpoint URL, done. Good for daily work in the editor you already have open.
OAuth for Claude Desktop and ChatGPT connectors. You add the MCP endpoint, your browser opens a MetriXs page, you log in, pick one site, and allow access. No key copying at all. Cursor and JetBrains support this flow too.
What the assistant can and cannot do
| Tool | What it returns |
|---|---|
| Overview and timeseries | Visitors, pageviews, bounce rate, duration, revenue per hour or day |
| Breakdowns | Top pages, sources, countries, devices, UTM parameters, custom events |
| Event properties | Property values per event, e.g. which plan was picked at signup |
| Commerce | Revenue, average order value, conversion rate, funnel, top products |
| AI referrals | Which AI assistants send you visitors, and where they land |
What it cannot do: change any data. There are no write tools. No settings changes, no site edits, no deletions. Every tool is a read-only query, and every response is aggregate numbers. Visitor-level identifiers do not exist in the responses at all, because MetriXs never stores them in the first place.
Who talks to whom
MetriXs itself never calls an AI provider. The data flow is your assistant asking MetriXs a question and MetriXs answering with numbers. No LLM runs on our side, no analytics data is sent to any AI company by us. You choose the assistant, you hold the credential, and you can end the connection at any time. Your analytics stay on the EU-hosted instance; the answers you asked for go to the chat you picked.
That also means there is no new subprocessor in the MetriXs privacy story: nothing changed about how visitor data is stored or protected. The only new surface is the one you deliberately connected yourself.
Getting started
- Claude Desktop or ChatGPT: add a custom MCP server (or connector) with the endpoint
https://mcp.metrixs.eu/api/mcp, then approve access for one site in the browser window that opens. - Claude Code, Codex, Gemini CLI, VS Code Copilot: create an API key under Settings → Sites and add the endpoint with a Bearer header to your client's MCP config.
The full setup steps per client, including ready-to-paste config snippets, are in the MCP guide.
Frequently asked questions
- What is an analytics MCP server?
- A server that speaks the Model Context Protocol and exposes analytics tools to AI assistants. Connect Claude, ChatGPT, Copilot, Gemini or Cursor to it once, then ask questions about your traffic in plain language and get answers based on your real data.
- Which AI assistants work with MetriXs MCP?
- Anything that supports remote MCP servers. API keys work with code assistants like Claude Code, OpenAI Codex, Gemini CLI, VS Code with GitHub Copilot, Cursor and JetBrains. The OAuth flow covers Claude Desktop and ChatGPT connectors.
- Is it safe to give an AI assistant access to my analytics?
- The connection is read-only and pinned to one site, and it shows up in your dashboard as a key you can revoke instantly. The assistant can query aggregates like visitor counts and top pages, but it can never change anything and never receives visitor-level identifiers.
- Does MetriXs send my data to an AI provider?
- No. MetriXs never calls an AI provider. Your assistant connects to MetriXs and pulls the answers itself; nothing is pushed to Claude, OpenAI, or anyone else by us.
- How do I connect Claude or ChatGPT to my MetriXs data?
- Add the MCP endpoint https://mcp.metrixs.eu/api/mcp as a custom server or connector. Your browser opens a MetriXs consent page where you log in and pick one site. After that, the assistant can read that site's analytics.
Ask your assistant about your traffic
First-party MCP server, read-only, one site per connection, EU-hosted. Included in every plan, free plan forever.
Set up the MCP server →Related: Measuring AI referrals · Generative Engine Optimization · Privacy-friendly web analytics · How cookieless tracking works