Why I threw away the frontend on my first product

My first product was called Cheap Analytics. The goal was simple: easy to set up, designed for people running multiple side projects, and cheap as chips. With a few little differences, it looked like every other analytics tool. The main distinction was in the name, and the one extra feature was a weekly AI summary email that told you what happened across your sites so you didn't have to log in.

That email was the thing people actually liked. A simple analysis of what was happening, shared where you were. The most useful part of the product was the part that meant you never had to open the product.

Around the same time, MCP started taking off and I was spending most of my day inside Claude Code. I wanted my AI analysis there, not just summarising your analytics once a week, but querying them whenever it needed to. Not a dashboard with an AI bolted on, but analytics where the AI is the interface.

So I started from scratch, rebuilding it as an API with 37 MCP tools optimised for how agents work, and giving it a new name. Traffic sources, page performance, date comparisons, geographic breakdowns, custom events, the lot. Your agent connects, queries what it needs, and when a human actually wants to see a chart, the agent generates one on the fly. Design for what an LLM needs: focused, structured, token-saving output.

There's still a website for signing up and managing API keys, but the core product has no GUI. Your agent is the interface.

A month in now. Free up to 2,500 events/month, €9.99/mo for 100K. Tracking script is 2KB, setup is about two minutes if you're already in Claude or Cursor.

This obviously only works if you already have an AI agent running. If you want to check your traffic the old-fashioned way, Plausible and Fathom are great. But if you're in Claude Code all day anyway and you'd rather your agent just knew your traffic without you alt-tabbing to anything, that's what this is.

https://lodd.dev if you want to try it.