Test coverage written and kept up by agents.
Three agents split the testing work: one plans scenarios from your app, one writes the Playwright tests, and one repairs tests a UI change has broken. When the feature itself is broken, the healer skips the test rather than forcing it to pass.
What it is
The MCP server inside Playwright Test that its three built-in test agents use. Playwright is Microsoft's tool for testing web apps in real browsers. You do not usually add this server by hand: Playwright's init-agents setup command writes the agent definitions for Claude Code, VS Code, Codex or OpenCode, and they call it. It is a different server from @playwright/mcp, Playwright's general browser-automation server.
What you get
- A planner that explores your app and writes a readable Markdown test plan
- A generator that turns the plan into Playwright test files, checking selectors and assertions live as it goes
- A healer that replays a failing test, finds what changed in the page, patches the test and reruns it until it passes
- Plans and tests stored in your repository, in specs/ and tests/ folders, where you can review them
Requirements
Node.js 20 or newer and a Playwright Test project. Run the init-agents command with the loop for your tool, for example claude for Claude Code. Run it again whenever you update Playwright, so the agents pick up new tools and instructions. In VS Code, version 1.105 or later is needed. No account or key.
Setup effort
One command — npx playwright init-agents --loop=claude