Data & MCP
Your AI assistant is guessing about the World Cup
Ask Claude or Cursor who reaches the final and it answers in a heartbeat, fluent, certain, and quietly inventing half of it. Training data that froze months ago. Squads that have turned over since. No calibrated probabilities, so it reaches for the shape of past tournaments and calls it analysis. The confidence is real. The numbers aren't.
Closing that gap takes one line of config.
Without the MCP server
Confident, fluent, and often wrong, the assistant improvises from training data that doesn't know this tournament.
With it
The same fluency, now answering from 157,119 live data points, updated after every match day through the World Cup, not remembered from years ago.
Hand it the real numbers
The hosted Model Context Protocol server gives your assistant the model output as tools it can call mid-conversation, 12 of them, one per dataset. Add it once. From then on, “compare France and England's projected XIs” or “how likely is Brazil to reach the semifinal?” is answered from the live model, not a guess.
157,119
data points your assistant can reach
13,605 live records across 12 datasets, every figure model-generated or drawn from the public record. Updated after every match day through the tournament.
One command
Add it to Claude or Cursor
Authenticate with an API key, any signed-in account can generate one, and your first 10 tool calls are free. The handshake that adds the server and lists its tools doesn't count; only the data calls do.
claude mcp add --transport http onthepitch \ https://onthepitch.now/api/mcp \ --header "Authorization: Bearer otp_live_…"
Then prompt naturally, “which team is most likely to win Group H?”, “compare the projected XIs for France and England”, “how did Brazil do at WC2022?”, and the assistant calls the matching tool (get_forecast, get_predicted_squads, get_majors) and answers from the live data, every tool mirrors a /api/v1/ endpoint, so REST and MCP can't drift.
Go deeper
Full data catalogue & API
All 12 endpoints as JSON or CSV, the bulk snapshot zip, historical-tournament slicing, and copy-paste pandas recipes for fitting your own model.
Open →
Interactive API reference
Every request/response schema, a try-it-live console, and the OpenAPI 3.0 spec at /api/v1/openapi.json.
Open →
Update log
What changed under the model, refits, ingestion changes, methodology tweaks, in chronological order, as it happens.
Open →
Try it free
10 free calls to try it
Sign in and generate an API key to get 10 free programmatic calls, shared across the REST API and the hosted MCP server, enough to wire the data into a notebook or an assistant and see it working. The MCP handshake doesn't count; only the data calls do.
Pro Pass
Go unmetered with the WC2026 Pass
$35 one-time, access through 31 Dec 2026. Unmetered API + MCP access updated after every match day, the bulk snapshot zip, historical tournament slicing, and the live update log, plus everything the Standard Pass includes.
Every forecast graded against the real result, scored on 987 matches since 2014. See the scorecard.
24h money-back, no questions asked·No subscription, no auto-renewal·Access through 31 Dec 2026. See refund policy.