mrkeyoor.comAsk

ccxt/ccxttested 03 Oct 2026commit 70b73bf

ccxt review

Our CCXT run installed 498 packages in 25 seconds, then both build and test stopped at an ESLint SyntaxError.

  • Build failed
  • Tests failed
25 sto install
  • 498 packages
Dataevaluationupdated 04 Oct 2026
+53stars / 7d

CCXT is an MIT-licensed library that gives applications one interface for market data and trading across more than 100 cryptocurrency exchanges and prediction markets. It saves developers from writing a separate connector for every venue, while leaving strategy, storage, and risk controls to the application.

Verdict

Our CCXT run installed 498 packages in 25 seconds, then both build and test stopped at an ESLint SyntaxError. Use the published package when broad exchange coverage is worth maintaining venue-specific regression tests. Source contributors should treat commit 70b73bf as a failing baseline, while trading teams should pin releases and validate every method that can move money.

We ran it

Lab card: what happened when we ran ccxt
Lab card from our sandbox run
Screenshot of ccxt (docs.ccxt.com)
docs.ccxt.com
Install✓ · 25s498 packages · 338 MB
Build✗ · 45s
Tests✗ · 39sran, no count parsed
Known vulns00 critical · 0 high · 0 moderate · 0 low (npm audit)
Repo9343 files~6,751,569 lines of source · 439.3 MB · 16 CI workflows · Dockerfile

Answers from our run

Does ccxt build from source?

Dependencies installed in 25 seconds (498 packages), and the build failed. We cloned commit 70b73bf into a clean Debian container with 3 CPUs and no project-specific setup.

Do ccxt's tests pass?

No. The project's own test command exited unsuccessfully after 39 seconds, and its output did not expose an individual test count.

Does ccxt have known vulnerabilities in its dependencies?

npm audit found none in the dependency tree at the time of our run.

Who should not use ccxt?

Anyone looking for a finished trading bot or backtesting product: CCXT supplies exchange connections, not strategy, portfolio logic, or deployment.

What are the alternatives to ccxt?

Freqtrade, Hummingbot, Cryptofeed. Our CCXT run installed 498 packages in 25 seconds, then both build and test stopped at an ESLint SyntaxError.

Setup3/5Install passed in 25 seconds; build and test both failed
Docs4/5Detailed manual, examples, CLI, rate limits, and MCP guidance
Community5/544,248 stars, an October 3 push, and active issue work
Maturity4/5Wide venue coverage, offset by a failing pinned source build

Who it’s for

Developers collecting prices, order books, trades, or candles from several exchanges.
Trading teams that want the same core API in JavaScript, Python, C#, PHP, Go, Java, or Rust.
Engineers building their own bots or backtests who need connectors rather than a finished strategy product.
MCP and Claude Code users who want public market data first and can restrict private account permissions carefully.

Who it’s NOT for

Anyone looking for a finished trading bot or backtesting product: CCXT supplies exchange connections, not strategy, portfolio logic, or deployment.
Contributors who require a green source baseline on Node 22: at commit 70b73bf, both our build and test commands stopped at the same ESLint SyntaxError.
Data teams that assume every OHLCV volume field has the same unit: open issue #30673 says the unit may be base, quote, or contracts and is not exposed in machine-readable metadata.
Low-latency WebSocket systems unwilling to run exchange-specific regression tests: open issue #30220 reproduces a buffered update being dropped between watch calls.
Desks unwilling to inspect fee options: the README says its listed builder-program exchanges add 1 bps unless builderFee is disabled.

Setup reality

Our Node 22 sandbox installed 498 packages in 25 seconds and used 338 MB. The build failed after 45 seconds, and the test command failed after 39 seconds. Both stopped in ESLint 9.39.4 with SyntaxError: Invalid regular expression flags; the log does not identify the offending expression or its cause.

Public market data needs no account or API key. Private balances and trading require exchange-issued credentials, sometimes after identity verification. The official MCP server starts with public data, while trading, withdrawals, and raw endpoints are separate opt-in capabilities.

The commit 70b73bf checkout held 9,343 files, about 6,751,569 source lines, and occupied 439.3 MB. It had 16 CI workflow files, a Dockerfile, and a compose file, though our scan found no tests directory. Npm audit reported 0 known vulnerabilities.

CCXT gives 104 exchange integrations one common vocabulary

CCXT's README lists 104 cryptocurrency exchange integrations and a separate set of 9 prediction-market APIs. Applications get common methods for tickers, order books, trades, candles, balances, and orders instead of maintaining a different client for every venue. That is useful: a team can add another market without redesigning its whole connector layer. CCXT remains library software, so it does not supply a trading strategy, portfolio rules, storage, or production monitoring.

Version 4.5.85 covers JavaScript, TypeScript, Python, C#, PHP, Go, Java, and Rust. Several implementations are generated from the TypeScript source, which explains part of the repository's size. Public REST calls work without an account. Private methods need exchange-issued credentials, and some exchanges require identity verification before providing them. WebSocket classes cover streamed data, while a CLI and local MCP server provide other ways to call the same exchange layer.

What happened when we ran it

Our sandbox installed commit 70b73bf in 25 seconds. Npm added 498 packages and the installed tree occupied 338 MB. The environment was a fresh, unprivileged Debian container with 3 CPUs, 8 GB of RAM, Node 22, and no secrets. Npm audit found 0 known vulnerabilities across critical, high, moderate, and low severity levels. Installation was the only main step that completed successfully.

The build failed after 45 seconds, and the test command failed after 39 seconds. Both paths reached ESLint 9.39.4 and stopped with SyntaxError: Invalid regular expression flags while loading an ES module. The supplied log tail does not name the expression, file, or underlying cause, so blaming Node, ESLint, or project code would be a guess. The test path never produced a completed test summary because it stopped at the same lint error.

The 439.3 MB source tree is contributor-sized

The checkout contained 9,343 files and about 6,751,569 lines of source before installation. It occupied 439.3 MB, then the npm dependencies added another 338 MB. Our scan found 16 CI workflow files, a Dockerfile, and a compose file, but no tests directory. That last signal describes the directory scan, not an absence of test code elsewhere in a repository whose build scripts and generated language targets span several layouts.

Package users have a much shorter path than contributors. The README shows one-command installs from npm, PyPI, Packagist, NuGet, Go modules, Maven, and crates.io. Docker is presented as a development option for the cloned repository, not a requirement for applications importing CCXT. Our 25-second npm install supports that consumer story. It does not rescue the source build, which failed before completion on the same commit.

One API still needs venue-level data checks

CCXT smooths over naming, authentication, and response shapes, but exchange differences remain observable in the common fields. Open issue #30673 on OHLCV volume units says candle volume may represent base currency, quote currency, or contracts depending on the exchange and market type. The report says that unit is not available as machine-readable feature metadata. A plausible number can therefore be wrong by 1,000 times after an incorrect conversion.

Streaming deserves the same caution. Open issue #30220 on buffered WebSocket updates reproduces a second message being dropped when 2 updates with the same message hash arrive in one event-loop turn. That report concerns CCXT Pro 4.5.77 rather than our sandbox build, so it is issue evidence, not our measurement. For balances, fills, or positions, write regression tests around the exact exchange, market type, and CCXT version you deploy.

Private calls add credentials, permissions, and fee choices

Public price data needs no key, while private APIs can place orders, transfer funds, and request withdrawals. The official MCP server starts with public data and keeps trading, withdrawals, and raw endpoints behind separate opt-in capability tiers. Its README describes confirmations, notional caps, and an audit journal. Those controls help, but an application still owns exchange permissions, secret storage, IP restrictions, and the decision to disable withdrawals on a trading key.

The README also names 7 builder-program exchanges and says calls through those integrations add 1 bps unless exchange.options['builderFee'] is set to false. That option belongs in a configuration test because it affects cost. Rate limits need similar attention: CCXT defaults to a leaky-bucket limiter and offers a rolling-window mode whose documented default window is 60 seconds. Neither setting guarantees that every venue will accept the same request pattern.

Version 4.5.85 arrived two days before our run

GitHub showed 44,248 stars and a last push on October 3, 2026. Release v4.5.85 shipped on October 1, one day before commit 70b73bf. The open queue held 240 issues and 472 pull requests when checked, which explains the repository API's combined count of 712. Those figures show a heavily used project with current work, not a quiet library living on old coverage claims.

The same v4.5.85 notes include fixes for WOO order cost, BingX history filters, Coinbase International candle calculations, Binance rate-limit costs, and several other venue details. Frequent repair is both a reason to use CCXT and a reason to pin it. The project absorbs exchange churn for you, yet each update can touch money-facing parsing and order behavior. Review release notes and replay your venue tests before upgrading.

Choose a bot platform when connectors are not the whole job

CCXT is the strongest fit when your application already owns strategy and operations but needs access to more than 100 venues. Freqtrade adds Python strategy development, backtesting, and bot operation. Hummingbot centers bot deployment and market-making workflows. Cryptofeed narrows the problem to normalized WebSocket market data, which is attractive when private trading methods would only widen your security surface.

Start with a published CCXT package, one exchange sandbox, and the smallest API permission set that works. Our run gives contributors a different instruction: do not call commit 70b73bf green on Node 22, because its build and test commands both stopped before completion. The abstraction saves connector work only if your own checks catch the venue-specific mistakes that a common method name cannot remove.

Alternatives

ProjectWhat it isPick it when
Freqtrade gh↗A Python crypto bot with strategy development, backtesting, optimization, and live operation.pick this instead when you want a working bot framework rather than an exchange connector library.
Hummingbot gh↗A platform for building and running automated crypto trading bots.pick this instead when market making and bot deployment matter more than embedding exchange APIs in your own application.
CryptofeedA Python library focused on normalized WebSocket market-data feeds.pick this instead when live market data is the job and you do not need CCXT's private trading surface.

Sources

  1. CCXT repository and README
  2. CCXT v4.5.85 release notes
  3. CCXT manual
  4. CCXT MCP server documentation
  5. Issue 30673: OHLCV volume units
  6. Issue 30220: buffered WebSocket updates

More data reviews