Playwright
E2E testing and browser automation. Screenshots, traces, multi-browser. Also works as the eyes of a coding agent: Claude Code opens the browser, screenshots the page it just built and spots what is broken before you look. For that job, install the command line the Playwright team ships separately as the `@playwright/cli` package, not the MCP server: its commands avoid loading large tool schemas and the page's full accessibility tree into the model's context, and `playwright-cli install --skills` installs the skills that describe them to your agent. Keep the MCP server for a long exploration or an autonomous flow where the agent has to hold the page state and reason over its structure from one step to the next. Your CI test suite does not change. The command line can also hold several browsers at once: each session has a name (`playwright-cli -s=name`), so the agent can run several flows side by side without them stepping on each other.
Strengths
- Native multi-browser: Chromium, Firefox, WebKit in parallel
- Smart auto-waiting: waits for element to be visible and interactable
- Codegen: record browser actions → auto-generate the test
- Trace viewer: test video replay + network + console, perfect debugging
Limitations
- E2E tests inherently slow, a full suite takes 5-30 min in CI
- Flakiness on dynamic apps with animations or random timings
- Not suitable for unit tests or pure logic testing
Best for
- Critical user flows: onboarding, payment, authentication
- Web scraping and repetitive browser task automation
- Visual check loop for an agent: it builds, looks at the render, fixes