Lumify

Lumify is a sports data and intelligence API built for AI agents. It serves schedules, live scores, odds from multiple sportsbooks, player props, player statistics, injuries and betting splits as deterministic JSON, and layers predictive intelligence on top, including vig-stripped probabilities, fair prices and expected value with a written rationale attached. It also publishes player props catalogs and forecasting, so props are a first-class object rather than an afterthought. It's reachable as a REST API, through TypeScript and Python SDKs, and as an MCP server, so an agent in Cursor or Claude Desktop can call it as a tool without custom glue.

The problem it addresses is that sports data was built for dashboards, not for agents. Odds feeds arrive in formats meant for humans, schedules and scores live in separate services, and turning a raw line into a probability an agent can reason about means doing the vig math yourself. Lumify packages the whole stack behind one key and one schema, and it treats the intelligence layer as part of the product rather than an exercise left to the caller.

Coverage at the time of writing is eight sports and seventeen leagues, with NFL, NBA, MLB, NHL, college football and basketball, soccer and tennis named on the site. Odds come from nine books, Pinnacle, FanDuel, DraftKings, BetMGM, Caesars, Bet365, Circa, Hard Rock and BetOnline, with line movement history and timestamps so an agent can see how a price got where it is. Endpoints cover events, odds, player props and intelligence, authentication is a bearer token, and the MCP server exposes 24 tools such as listing events, fetching odds and pulling intelligence, with one-click installs for Cursor, VS Code and Claude Desktop and a Postman collection. Browser-based connectors for ChatGPT and Claude.ai are noted as waiting on OAuth support that isn't implemented yet.

The intelligence layer is what the name is pointing at. For a given market it returns a vig-stripped probability, a fair price, and a price overlay that shows where each book sits against that fair price, with a written rationale embedded in the response so the reasoning travels with the number. Injuries and betting splits feed into the same picture, and the line movement history lets an agent distinguish a price that has drifted for hours from one that just jumped. None of this replaces a model of your own, but it gives an agent a defensible starting point without a maths library.

It's for developers building automated betting agents, market anomaly detection, agentic sports media and ordinary sports apps that want structured data. The site talks directly to coding assistants as well as to people, and the free trial key that works without signup is aimed at an agent that wants to try a call before a human fills in a form. Getting started is a package install. The TypeScript SDK is on npm as @lumifyai/sdk, the Python SDK is on pip, and the MCP server can be added to Cursor, VS Code or Claude Desktop from the site with one click. Documentation, a changelog and free cost estimates through an estimate endpoint round out the developer surface, and the estimate call means an agent can check what a query will cost before it spends anything.

What makes it different from a generic odds feed is the agent-first framing. Deterministic schemas mean the same query returns the same shape every time, which matters when a model is parsing the output. The intelligence endpoints return probability and pricing with embedded rationale, so an agent can quote why a line looks mispriced instead of just that it does. And credits are only charged for successful responses, with errors and unavailable data never billed, which is the kind of detail that shows someone has run an agent against a metered API before and been annoyed by the bill. The free tier's 1,000 credits don't expire either, so there's no pressure to burn them in a month.

Pricing is freemium with a metered ladder. The free tier gives 1,000 credits that never expire, full access to the intelligence API, two API keys and 20 requests a minute with no credit card. Pay as you go is $5 per 1,000 credits at 60 requests a minute and five keys, aimed at under 40,000 credits a month. Growth is $199 a month for 50,000 credits with overage at $4 per thousand, 120 requests a minute and 20 keys, which the site puts at about 20 percent cheaper than pay as you go. Enterprise is custom, with unlimited credits and keys, a 99.9 percent uptime SLA, a dedicated support channel and custom endpoints. A standard call costs one credit. There's no published email, so contact runs through the site's form.