Why I built this
For a long time, picking a trail race in Catalunya was harder than running one. There were no real filters and never enough information, so every time I'd find a race and then do all the work by hand — how far is it to drive, how much climbing, is it even the kind of race I want. Every single time. It should be the easy part, and it was the opposite.
Because the races are worth it
There are a lot of them, but only 20 or 30 are truly special — and those are one of a kind. Some take you through the most beautiful places in Catalunya; I love running the trails on Montserrat, and a lot of these races are really just an excuse to explore the national parks and corners of this region you'd never see otherwise. Others are simpler — you run, and at the finish there's butifarra and a cold beer waiting. I love mountain races, and no two are the same.
My favorite is the Burriac Atac — great atmosphere, a proper race, real support on course, good food after, and even the start pack is done right. That's the whole package, and it's what I'm always chasing.
Why an AI, not just filters
Here's the honest problem: I can't tell you what filters you need. Distance, drive time, elevation — sure. But “a race with a great atmosphere and good food after,” “somewhere I've never been,” “something my kid can do too” — those aren't checkboxes. That's what pulled me to LLMs. You can ask an assistant the questions that don't fit a filter, and it just answers. To me that's the most personal search there is, and I wanted to play with it.
Built for agents, on purpose
So the whole thing is built to be used by an AI, not just read by a person. The race database is exposed as a live MCP server— a connector you can add to your own Claude or ChatGPT. You ask “find me a scenic 20 km race under an hour from Barcelona in October,” and it queries the live data and plans for you. No copy-pasting, no stale screenshots.
And because it's your assistant doing the work, it can combine this with things I'll never see. Connect a Strava or Garmin MCP alongside it and your AI can tell you whether you're ready for a given race and roughly what time you'd run — joining your training with these races locally, in your session. Your training data never touches my server.
What it is
Honestly, a pet project. I built it for me first, but I want it to be useful to anyone trying to find and explore races here — and it's my way of iterating on what a running marketplace could actually feel like.
Kept honest
The data is re-scraped every week from the source calendars, so dates and new races stay current. It's not perfect — organizers change things, and some facts (start times, whether registration is open) are volatile. That's why the site and the agent both tell you to verify a shortlisted race at its official page before you count on it.
— Dima