Opening Google Analytics to answer a single question is a special kind of tax. The number you need is in there — you know it, because the dashboard told you so last week. But retrieving it means four clicks, two date pickers, and one dimension dropdown, and by the time you've fought through the UI you've forgotten what you wanted it for. A developer going by freema decided that tax was unacceptable and built Metrifyr: a remote MCP server that hands Claude read access to their Google marketing stack.

The MCP Angle

Metrifyr rides the Model Context Protocol, the open standard that lets LLMs reach outside their training data into live tools. What's notable here isn't just another connector — it's the architecture choice of a remote server rather than a local plugin. That means Claude doesn't need your credentials or a local runtime; the MCP server brokers the session and exposes only what the model is allowed to see.

Read-Only by Design

The read-only posture matters for anyone who has watched an agent hallucinate its way into a destructive action. Metrifyr deliberately scopes Claude to querying, not mutating — so the worst failure mode is a wrong answer about session counts, not an accidental campaign pause or budget overwrite. For marketing teams skittish about letting an LLM near their ad spend, that boundary is the whole pitch.

The Workflow Shift

The practical payoff is conversational analytics: instead of memorizing GA navigation paths, you ask questions and get answers grounded in live data. It's a small demo of a much bigger pattern — treating your data stack as something an LLM can interrogate directly rather than something you must translate through dashboards.

Key Takeaways

  • Metrifyr is a remote MCP server giving Claude read-only access to the Google marketing stack, replacing multi-step dashboard navigation with natural-language queries.
  • The read-only scope limits blast radius — hallucinations produce wrong answers, not destructive campaign changes.
  • Remote architecture keeps credentials out of local runtimes and centralizes what data the model can reach.

The Bottom Line

The source post is light on implementation specifics — setup steps, auth flow, and query latency are all left to the reader's imagination. But as a proof point for MCP adoption beyond toy demos, Metrifyr makes a compelling case that your analytics platform should answer questions in plain English by default. If you want details on how freema wired up OAuth scopes or which GA endpoints are exposed, you'll need to ask the author directly — this write-up is more manifesto than manual.