Skip to content

Introduction ​

bash
npx monkeytest run https://example.com --depth basic

You give it a URL. It opens Chromium with Playwright, decides what a person might try, runs those cases, and writes a report. A flow counts as passed only when an outcome check verifies. After you fix something, monkeytest rerun replays the same trace and monkeytest diff tells you what changed.

The package is @monkeytestai/monkeytester. The binary is monkeytest. It needs Node.js 20 or newer. The license is AGPL-3.0-only.

npm dist-tag latest is the stable channel. npm dist-tag next is the pre-release channel. These pages describe this repository. The engine version is package.json, not the hosted product. Channels, the 8 Nov 2026 registry observation, and the hosted 0.7.0 service are separated in Versions.

What a run is ​

Four phases, then a file you can keep:

  1. Explore. Crawl up to a page cap. Same-site links only (www and the apex match, and so do http and https).
  2. Plan. An LLM proposes flows from what the crawl actually saw. Page checks and auth-negative checks are added by the engine and do not spend that flow budget.
  3. Run. Playwright performs the steps. Screenshots land in .monkeytest/shots/.
  4. Triage. Confirmed site bugs are fingerprinted. Playwright's own misses stay inconclusive.

RunResult.outcome is bugs_found, clean, inconclusive, or could_not_test.

What it is not ​

It does not replace a script you already trust for a checkout or a login. Playwright Test is the framework for those tests, and Playwright Test Agents plan, generate, and heal them. monkeytester looks for cases the script did not name, then stores a plan you can rerun. It is not a hosted REST API and it does not ship an MCP server. See How is this different?.

It is not a load tester. axe-core runs as one page check (WCAG A/AA). That check is not a full accessibility audit.

Next: Quick start.

AGPL-3.0-only