CCXT

What it is

CCXT is a unified trading API. You write one integration and it talks to any supported exchange, mapping each venue's idiosyncratic REST and WebSocket surface onto one vocabulary - fetchTicker, fetchOrderBook, createOrder, cancelOrder, fetchBalance, fetchPositions. It ships in JavaScript and TypeScript, Python, PHP, C#, Go, Java and Rust from a single generated codebase, and it is MIT-licensed with nothing held back.

Most of CCXT is not about this catalogue's subject, and that should be said plainly. The library carries 104 crypto-exchange classes, 76 of them with WebSocket support. Prediction markets are a separate prediction namespace holding seven venues. If you install CCXT for event contracts, you are installing a large crypto library to reach a small corner of it, and the issue tracker, the release notes and the documentation you will be reading are overwhelmingly about something else.

It earns its card anyway, because it is the only client here that speaks to more than one prediction venue, and because the seven it covers are not toys - each implements order placement, cancellation, positions and order books, and all but one of them candles.

Availability

The package installs from PyPI, npm and the rest anywhere, and needs Python 3.10 or newer on the Python side. Public market data needs no key at all - you can read Polymarket's and Kalshi's order books with no account.

The flags here describe the library. Whether you may hold a funded position on any given venue is that venue's question, answered on that venue's card, and CCXT will happily construct an authenticated client for an exchange that will not serve you.

Sandbox mode is wired for some venues and not others. Among the prediction venues, CCXT carries test hosts for Kalshi's demo environment, Myriad and Hyperliquid; Polymarket, Limitless, Opinion and Binance prediction have no test URLs in their classes, so there is nothing to practise against.

Pricing

Free and MIT-licensed, including the WebSocket half - what used to be sold as CCXT Pro is now a free part of the library. There is no paid tier, no key, and no metered endpoint. The project takes sponsorships through Open Collective and carries paid promotion slots in its README, neither of which changes what the code costs.

Exchange fees are the exchange's. CCXT reads them back where the venue exposes them - fetchTradingFee is implemented for Polymarket and Myriad among the prediction venues - but charges nothing itself.

Markets & resolution

CCXT resolves nothing and settles nothing. It is a transport, and the settlement type on this card is the majority case among the venues it reaches rather than a property of the library - five of the seven settle in stablecoin on-chain, while Kalshi settles in US dollars as a regulated exchange. The same caveat applies to the liquidity model - most of these venues run an order book, but Myriad is priced against a curve, and fetchTradeQuote exists precisely because an executable quote on an automated market maker is not the same object as a book.

The data model is the part worth reading before you write anything. CCXT models prediction venues in three levels - an event is the question, a market hangs off the event, and an outcome (a YES or NO token, handled as something like TRUMP_WINS:YES) is the tradeable unit. Every price and trade method takes an outcome handle where a crypto method would take a symbol. Prices are probabilities between 0 and 1, amount is a number of shares, and cost is the collateral spent.

Discovery starts with fetchEvents, which takes a query, tags, an event id or a slug. Note one sharp edge - calling fetchTickers() with no arguments raises ArgumentsRequired rather than quietly returning a subset, because the venue has no all-tickers endpoint. Only Hyperliquid serves its whole universe in one request; the other six all raise, though the documentation names only four of them.

Integrations

The seven prediction venues, as of 2026-09-19 - Polymarket, Kalshi, Limitless, Myriad, Opinion, Binance's prediction markets and Hyperliquid's. Three of them have cards here for the underlying APIs, at polymarket-clob-api, kalshi-api and limitless-api.

WebSocket streaming is implemented inside the prediction classes themselves rather than in a separate Pro package, and only for three venues - Polymarket carries five watch methods over the CLOB market and user channels, Myriad eight, Opinion five. Kalshi, Limitless, Binance and Hyperliquid have none in their prediction classes, so anything live on those four means polling or your own socket code.

CCXT also ships an official MCP server, published as ccxt-mcp and run locally over stdio, which exposes market data, balances and opt-in trading to an MCP host. Keys stay on the machine; trading, withdrawals and raw endpoints are separate opt-in tiers that are off by default. Every language binding has a CLI as well, and the prediction namespace is reachable from it with a -p or --prediction flag.

Limitations

You are reading somebody else's translation of the exchange. That is the whole trade a unified API makes. When a field's meaning differs between Polymarket and Kalshi, CCXT has picked one mapping, and the error you get back is CCXT's rendering of the venue's error. For a single venue, the venue's own SDK or its raw API will always be closer to the truth.

The prediction support is newer and thinner than the crypto support. 76 of 104 crypto exchanges have WebSocket; three of seven prediction venues do. Capability varies enough between the seven that exchange.has[methodName] is not optional reading - Kalshi has editOrder and fetchSettlements, which Polymarket does not, while Polymarket has batch createOrders, fetchOrderTrades and fetchTradingFee, which Kalshi does not.

Python gets the async class or nothing. ccxt.prediction.<id> is the asynchronous class; there is no synchronous variant the way there is for crypto exchanges. Code that is otherwise synchronous has to grow an event loop to reach an event venue.

Documentation lags the code. The prediction-markets page currently names five venues in its opening sentence while the repository's own exchanges.json and README table both list seven. The machine-readable list is better on that count and worse on another: the predictionWs array in exchanges.json is empty, although three of the prediction classes do carry watch methods. Read the classes.

Size. 236 issues are open against the repository, and roughly 500 more open items are pull requests. Releases land about weekly - 39 of them in the six months to 2026-09-19, when 4.5.81 went out - which is a strength for a library tracking a hundred moving APIs and a cost if you pin versions carefully. Note that the npm and PyPI releases are not in step: PyPI had 4.5.81 on 2026-09-19 while npm was still on 4.5.78 from 2026-09-07.

Alternatives

For one venue and nothing else, a single-venue client is a smaller dependency and a shorter path to the exchange's actual semantics. Polymarket's is now polymarket-client, from the repository that replaced py-clob-client; Kalshi's is kalshi-python, which has not shipped since September 2025.

For reading prices without any client library, the venue REST APIs are plain HTTP - see polymarket-gamma-api, kalshi-api and limitless-api.

For prices from several venues in one response without writing the fan-out yourself, a cross-venue data provider such as adjacent or finfeedapi is a different answer to a neighbouring question - data only, no order path.