Trail Catalunya for AI agents
A live, read-only MCP server for every trail race in Catalunya. Connect it to your own Claude or ChatGPT and plan races off current data.
What is this?
Trail Catalunya publishes 200+ trail-running races in Catalunya as a Model Context Protocol (MCP) server. Any MCP-capable assistant — Claude (any plan, including Free with a single custom connector) or ChatGPT on a paid plan that supports custom connectors — can add it as a connector and query the live race database directly, instead of reading a static web page.
What is the server endpoint?
It's a public, no-auth, read-only HTTP MCP server (JSON-RPC, protocol version 2025-03-26). No API key needed. Queries are logged anonymously — no IP, no identity — for 90 days to improve the tool.
The homepage “ask” box also logs what people type, to learn what runners want. It's unlinked structured event data — not tied to any identity in our systems — and the free text you type is deleted after 90 days (the anonymous chip/filter counts are kept). It's read only by the site owner and used only as a directional hint about demand.
How do I connect it?
- Claude: Settings → Connectors → Add custom connector → paste the URL above. On Team/Enterprise an Owner must add it first under Organization Settings → Connectors.
- ChatGPT: turn on Developer mode in settings (under Apps/Connectors), add a custom MCP server, and paste the URL above.
The homepagealso has an “Ask AI” button with the URL and copy-paste steps, plus a zero-setup path that sends the currently filtered races straight into a chat.
What tools does it expose?
search_races— filter races by drive time from Barcelona, distance, elevation gain, month, province, and kids-run availability.get_race— full detail for one race: all distances, official URL, drive time from Barcelona, and data freshness.whats_on— the soonest upcoming races in a given window.
Each race carries its distances (km + elevation gain in metres D+), and most carry a real driving estimate from Plaça Glòries in Barcelona (not the user's location) — a few newer towns show no drive time yet — plus province, dates, a difficulty on ITRA's km-effort scale (km + D+/100; a 6-level word plus ITRA points 0-6, an endurance-load measure, not steepness — with a separate D+/km figure for verticality), and best-effort enriched facts. Many also carry a taste layer (what makes a race special, who it suits, its character), every field labelled by how we know it — organizer, derived, our read, inference, or Dima (ran it) — so an agent never mistakes our judgement for an organizer fact.
Can it use my training data?
Yes — locally, in your session. If you also have a Strava or Garmin MCP connected to the same assistant, it can join your recent training with these races and add three things: race readiness (are you prepared for the distance and climb), a rough projected finish time, and a suitability ranking by fitness fit. Ask “Am I ready for this race, and what time would I run?”
This join happens entirely inside your assistant. Your training data is never sent to this server — the server only takes race filters. If you don't have a training connector, add one to unlock this; without it, the assistant answers from race facts alone and won't guess.
Where does the data come from, and how fresh is it?
Races are re-scraped weekly from the Catalan trail-race calendars. Distances, elevation and dates are stable; volatile facts (start time, registration status, sold-out) should be verified at the race's official URL before you rely on them — the tools say so, and so should you.