Lumify Review 2026

Lumify is a sports data and odds API that treats AI agents as the primary consumer rather than an afterthought. It serves schedules, live scores, team and player data, odds from 9 sportsbooks, public betting splits, and a predictive layer the company calls bet intelligence: probability, vig-stripped fair price, and a structured rationale for each output.

The framing matters because it drives the product decisions. There is a hosted MCP server at a single URL with 23 tools. There is a free cost-estimate endpoint so an agent can price a call before spending on it. There is a trial key you can pull with no signup at all. Every metered response carries _meta.credits_used, so an agent can track its own budget. Most odds APIs bolt an MCP wrapper onto a REST product. Lumify built the metering and the discovery surface around the agent case first.

One note on the name, because it causes real confusion in search: this Lumify is not connected to LUMIFY eye drops, Philips Lumify ultrasound, or the unrelated lumifyai.com. The company disambiguates it in its own docs, which tells you how often people land in the wrong place.

Eight sports: NFL, NCAA Football, NBA, NCAA Basketball, NHL, MLB, tennis (ATP/WTA), and soccer across MLS, the Premier League, La Liga, Serie A, Bundesliga, Ligue 1, and the Champions League.

Book coverage is where you need to set expectations. Lumify carries 9 sportsbooks: Pinnacle, FanDuel, DraftKings, BetMGM, Caesars, Bet365, Circa, Hard Rock, and BetOnline. That is a curated US-facing list, not the 85 to 200+ book counts that aggregators advertise. Having Pinnacle and Circa in a 9-book set is a deliberate choice, since those are the two books most models want as a sharp reference. But if your product is a line-shopping tool that lives on breadth, nine books will not carry it.

Markets stay on the main lines: moneyline, spread, and totals, with three-way moneyline for soccer, and alternate spread and total rungs behind include_alts=true. Player props exist as a separate endpoint for NFL, NCAAF, NBA, NCAAB, NHL, and MLB; other sports return a 400. Futures are not on the public v1 surface at all. Same-game parlay pricing and sportsbook deeplinks are absent.

The predictive layer is uneven by design and the docs say so plainly. Full bet intelligence is live for MLB, NFL, NCAAF, tennis, and soccer (MLS plus the big five leagues). NBA, NHL, and NCAAB get data, odds, and player-prop forecasts but no bet objects yet. Champions League returns available: false.

This is the reason to look at Lumify. The MCP server is hosted, at https://lumify.ai/mcp, as a stateless Streamable-HTTP endpoint with 23 tools: list_events, get_odds, get_player_props, get_intelligence, estimate_cost, and the rest. No npx, no local process, no stdio bridge to babysit. Most providers on our sports MCP servers page ship a local package you install; a hosted URL you can point a remote client at is still rare.

The metering is the clever part. initialize, tools/list, and ping are free, so discovery costs nothing. Only tools/call bills, at the same rate as REST. Pair that with the free POST /v1/estimate endpoint and an agent can plan a query, price it, and decide, without a human in the loop. There is also an /api/agent/* surface for provisioning keys and topping up credits programmatically.

Around the MCP server sits a full set of integration packages: @lumifyai/sdk for TypeScript, lumify-sdk for Python, plus published connectors for LangChain, LangGraph, LlamaIndex, CrewAI, and n8n. The site also publishes llms.txt, a full OpenAPI spec, and a SKILL.md, with measured token counts for each so an agent can pick the context file that fits its window. That is a level of care most API docs do not bother with.

One limitation worth knowing up front: there is no OAuth yet, which means the ChatGPT and Claude.ai web connectors will not work. Desktop and IDE MCP clients do.

Scores update roughly every minute. Odds update roughly every 10 minutes, pregame and in-play alike. The docs state this outright and add that the product is "not suitable for sub-second trading." Believe them. A 10-minute in-play cadence is fine for a research agent, a splits monitor, or a scoreboard. It is not fine for live arbitrage or anything that races a book to a stale line.

Push delivery is SSE streams plus signed webhooks for score, status, line-move, and intelligence events, and both are available on self-serve tiers rather than locked behind an enterprise contract. There is no WebSocket, which is why the directory marks that column red. If your architecture assumes a socket, you will be writing an SSE client instead.

Two more edges to plan around. Historical queries are capped at 90 days per request. And final odds carry a result field graded won/lost/push/void, but graded from the pre-kickoff closing line rather than the last in-play price. That is settlement-adjacent data on main lines only, not a full bet-settlement product, so we mark that column as unverified pending confirmation from the team.

Four tiers, all published:

• Instant trial, $0: 100 credits, 14-day expiry, no signup required. You copy a key off the docs page and start.
• Free, $0: 1,000 credits that never expire, full access to every endpoint, up to 2 API keys, 20 requests per minute.
• Pay As You Go, $5 per 1,000 credits: no monthly fee, 60 requests per minute.
• Growth, $199/mo: 50,000 credits included, overage at $4 per 1,000, 120 requests per minute.
• Enterprise: custom, with a 99.9% uptime SLA.

Most successful calls cost 1 credit, including multi-book odds pulls and compound requests that bundle odds and intelligence together. Errors and available: false responses are free, which removes the usual annoyance of paying to discover that coverage is missing.

The free tier is one of the better ones in the free odds API field. Credits that never expire mean a hobby project can idle for months and still have its budget intact, and 1,000 calls is enough to build something real rather than just print a JSON blob once. The no-signup trial is the most developer-friendly onboarding we have seen from any provider in the directory.

The product maps cleanly to a few builds:

• Agent developers who want a sports tool their LLM can discover, price, and call without a custom integration. The hosted MCP plus free discovery calls is the whole pitch, and it lands.
• Research and analysis tools that need rationale, not just numbers. The intelligence layer returns confidence scores and signal breakdowns you can show a user.
• Splits monitors tracking ticket-percentage against handle-percentage divergence for MLB, NBA, NHL, and NFL.
• Scoreboards and score bots on the ~1-minute score cadence, with SSE or webhooks doing the push. There is an open-source demo in the company's GitHub repo.
• Prototypes and side projects that need to start free and stay free while they find a shape.

It is the wrong tool if you need book breadth, futures markets, same-game parlay pricing, deeplinks, or sub-minute odds. Those are all explicit gaps, not things we inferred.

Against the broad aggregators in the odds API directory, Lumify trades market coverage for agent ergonomics. A provider with 85 books and full prop depth will beat it on any question that starts with "how many." Lumify wins on the questions that start with "how quickly can my agent use this."

Credit to the company for arguing this honestly in its own docs, where it names its competitors and concedes the breadth gap rather than papering over it. That candour is unusual and it makes the technical claims easier to trust, though it is still marketing and you should verify against your own workload.

The realistic pattern for a serious product is two providers: Lumify for the agent interface and the intelligence layer, a breadth-focused feed for the long tail of books and markets. At $0 to start and $199/mo at the Growth tier, adding it alongside something else is not an expensive experiment.

Lumify is the most agent-native sports API in the directory right now. The hosted MCP endpoint, free discovery and cost-estimate calls, per-response credit metering, and no-signup trial key add up to a product someone actually thought through for LLM consumption rather than retrofitted.

The constraints are real and clearly stated: 9 books, main lines only, no futures, no WebSocket, and a ~10-minute odds cadence that rules out fast trading. Whether that matters depends entirely on what you are building. If you are wiring a sports tool into an agent framework, start here, because the free tier costs you nothing but an afternoon. If you are building a line-shopping product that lives on book count, look elsewhere first and come back for the MCP layer later.