How sportsdata works

The architecture in three steps, and every common question answered. The short version: live data in, grounded answers out, nothing invented.

How it works

Connect any LLM

Bring your own Claude, GPT or Gemini key — or plug the data plane straight into your own client as an MCP server.

Agents query live data

Specialist agents pull from 60 providers — bookmakers, prediction markets, official league APIs, racing & fantasy feeds — through one capability-tagged tool surface.

Grounded answers

Every number traces to a tool result, verified before it reaches you. Budgets cap every run. No invented facts, ever.

Reading the boards

Both terminals are live — the racing board follows money flow, the sports board measures every bookmaker against the sharp line.

The sharp line

The sharp price is a de-vigged blend of whichever of Kalshi, Polymarket, Betfair and Pinnacle priced the game — two prediction markets, an exchange, and the low-margin book that welcomes winners. All four are disciplined by money in a way a retail promotional price is not. De-vigging strips the bookmaker margin so implied probabilities sum to 100% and become comparable.

Value

Value is the gap between the best price any book offers and the sharp fair price for the same outcome — a statement about disagreement between markets, not a prediction about the game. Markets no sharp has priced still appear with book-vs-book comparison.

Same-game multis

The SGM panel quotes combinations through the bookmakers themselves — real, bookable prices carrying the book's own correlation adjustment — with a model-free independent price as the floor.

Racing money flow

The racing board tracks which runners shorten as money arrives across tote, exchange and books, with a per-race pick and a fair price for every runner.

Informational only. Nothing here is betting advice; 18+.

Questions, answered

What is an MCP server for sports data?
The Model Context Protocol (MCP) is an open standard that lets an AI client call external tools. sportsdata is an MCP server that exposes live sports data and bookmaker odds as tools, so an assistant like Claude Desktop or Cursor can fetch real fixtures, stats and prices instead of guessing from memory.
How much does sportsdata cost?
Nothing. Both the MCP data plane and the agent workbench app are free and MIT licensed. There is no account and no API key to buy. If you use the agent app you supply your own model key, and you pay your model provider directly for that usage. Donations are voluntary and unlock nothing.
Which sports and bookmakers are covered?
Over 60 providers. Sport and stats feeds cover AFL, NRL, NBA, MLB, NFL, NHL, EPL, cricket, golf, tennis, F1, UFC, and horse, harness and greyhound racing. Odds come from bookmakers including Sportsbet, TAB, Pinnacle, Unibet, PointsBet, Ladbrokes, Neds, betr, Dabble and FanDuel, plus odds aggregators and the Kalshi and Polymarket prediction markets. Browse every tool.
How do I install it in Claude Desktop or Cursor?
The server is published on PyPI, so a single line wires it into any MCP client: uvx sportsdata-mcp serve. There is also a prebuilt app that needs no Python — download it, unzip, and run ./sportsdata-mcp setup. See Get it.
Do I need my own API keys?
For most providers, no. A subset of feeds require a key you obtain yourself from that provider — those are marked with a 🔑 in the tool browser above. Everything else works with no credentials at all.
Does sportsdata give betting advice?
No. It returns data — prices, stats, fixtures and model outputs — for information only. Nothing on this site or in the tools is betting or financial advice, and the service is for adults aged 18 and over. Gamble responsibly.
Is my data sent anywhere?
No. The MCP server runs on your own machine and talks directly to the upstream data providers, so your conversations and your model key never reach us. This website uses Cloudflare Web Analytics for page counts, which sets no cookies, does no fingerprinting and does not track you across other sites — see the privacy policy.