Home / Directory / Playwright MCP

Playwright MCP: let the agent drive a browser through the accessibility tree

MCPCURSORCLAUDE CODEANY STACK

By Microsoft · License: Apache-2.0

This is Microsoft's MCP server for Playwright. It is a real browser tool, and the README is unusually honest about when you should not use MCP at all.

Open Playwright MCP

What it does

The server lets an MCP client launch Playwright and operate pages via the accessibility tree: navigate, click, type, inspect. No vision model required. Typical install is npx @playwright/mcp@latest in the client's MCP config.

The same README now compares MCP to Playwright's CLI + skills. For coding agents, CLI calls can be cheaper because they avoid dumping large tool schemas and verbose snapshots into context. MCP still fits long, stateful browser loops.

Requirements are ordinary: Node 18+, any MCP client (Cursor, Claude Desktop, VS Code, Windsurf, and others listed in the repo).

Who it is for

People writing or repairing end-to-end tests, or agents that must click a real UI. Also useful for "does this page actually render" checks after a frontend change.

If you only want the agent to edit code and run npx playwright test, you may not need this server. A skill or a documented command in AGENTS.md can be enough.

Our take

First-party Playwright, Apache 2.0, and a design that avoids screenshot-only agents. That combination is why this server shows up in every MCP list.

Limits: it can change the world (it drives a browser). It is not a rules pack. Token cost is real if you leave it connected all day. Microsoft themselves suggest CLI + skills for many coding sessions.

We are not quoting download counts. The repo is the source; pin a version in production instead of @latest if you care about reproducibility.

Compared with AgentKit Drawer

AgentKit Drawer does not ship a browser MCP. Our kits tell the agent when to run tests and what "done" means. Playwright MCP is how the agent becomes the test runner.

No related-pack button: none of the drawers are a Playwright setup. Put the test command in AGENTS.md, and add this server only if the agent must operate the UI itself.

FREE · 1 FILE · WORKS IN EVERY AGENT

Try the free Starter AGENTS.md first

A 40-line rules file that makes Cursor, Claude Code, Codex and Windsurf plan first, keep diffs small and run your tests before saying "done". It works alongside anything in this directory. We email it to you instantly.