Horse Truth API Review 2026

Horse Truth rates Australian thoroughbreds, most of them racing in Queensland, and sells that rating to software. You ask about a horse and get its type, readiness, form trend, best distance and a short list of signals, each one written as a plain sentence. The same answers come through a REST API and a remote MCP server, so an AI agent can call it as easily as a script can.

The key word on every response is derived. Horse Truth does not resell race records, results or anyone else's data. It sends its own reading of that data, and every payload says so with fields like derived_output_only and raw_provider_data_included: false. It also uses no odds. That makes it a ratings service, not a price feed.

The free sandbox is the quickest way to see the data. It takes any horse name:

curl "https://horsetruth.com.au/api/v1/sandbox/horse?name=Final%20Crusade"

We ran that call. The teaser came back with the horse's type ("Class horse", with a one-line explanation), a best distance of 1300m, a form trend of "Holding", five runs analysed and an evidence depth of "Low". Two signals followed. One was a running profile (even pace over its last six runs). The other came from race tracking data: the horse lifted its stride rate late, the best of the 11 runners at Eagle Farm on 10 October.

The sandbox shows one signal of each kind and tells you how many it held back. Each teaser also lists the paid call that returns the full set, with its price, so a developer or an agent always knows the next step.

The evidence depth field is worth a look. A rating built on five runs is a different thing from one built on thirty, and Horse Truth says which one you have instead of hiding it.

The paid REST endpoints cover the full horse read (/horses/{horse}/intelligence), material changes over time (/horses/{horse}/changes), name resolution, class rankings, side-by-side comparison of two to five horses, and a full read of one race with every runner ranked. A watchlist endpoint lets you follow horses and poll a feed of change events for them. An OpenAPI 3.0 spec and a Postman collection cover all of it.

The MCP server sits at https://horsetruth.com.au/api/v1/mcp and is listed in the official MCP Registry. It has eleven tools. Two are free: a sandbox preview and a guide to the purchase options. The other nine cost one credit each. We called tools/list on it with no key and got all eleven back. Our sports MCP servers guide lists it with the other official servers.

Authentication is a key in the Authorization: Bearer header or in X-Horse-Truth-Key. Send an Idempotency-Key header on paid calls: a retry with the same key does not use a second credit.

There are three ways to pay.

• Credit packs and monthly plans. A one-time pack of 50 credits costs A$1, and 250 credits cost A$5. Monthly plans start at A$49 for 5,000 credits and go up to A$999 for 300,000. One successful call uses one credit. Monthly plans bill overage through Stripe when the credits run out; packs stop with an HTTP 402 instead.
• x402, per call. x402 is a pay-per-call standard that settles each request in USDC, a dollar stablecoin, on the Base network. It needs no account. Resolving a horse name costs US$0.001, and a full read costs US$0.02. This route suits an autonomous agent that holds its own wallet.
• RapidAPI. Horse Truth is also on RapidAPI, which handles the keys and the billing if your team already buys APIs there.

The free sandbox stays free: no key, 30 requests an hour per IP.

• Odds. None, by design. For Australian racing prices, the directory lists odds feeds that carry thoroughbred, harness and greyhound markets.
• Raw race data. No results tables, sectional times or form guides for you to model yourself. You get Horse Truth's reading of them.
• Coverage outside Australia. The ratings are Australian, and the depth is greatest in Queensland.
• Harness and greyhounds. Thoroughbreds only.
• Push delivery. The change-event feed is polled. There is no WebSocket.

• AI agents and chat assistants that need to answer "how is this horse going?" with a sourced, plain-English answer. The MCP server and x402 billing are built for exactly that.
• Racing content sites and newsletters that want a readable verdict on a runner without building a ratings model.
• Owners, syndicates and watchlist apps that care about change: a horse whose readiness or form trend moves is the event worth an alert.
• Odds tools that already carry Australian racing prices and want an independent rating next to them.

If you want the raw data to build your own ratings, look at our sports data API guide instead.

This is a narrow API with a clear job. It turns Australian racing data into short, honest ratings and sells them by the call to software and agents. The free sandbox works on any horse with no key, the MCP server answers out of the box, and the A$1 pack makes a real test nearly free. It also publishes a sealed public record of its race-day selections, so you can check how its calls have gone before you lean on them.

Do not expect prices or raw data. Expect a rating with its evidence depth stated. Run the sandbox on a horse you know and read the developer docs for the full endpoint list.

Last updated: 15 October 2026 · How providers get listed