Playwright MCP
A Model Context Protocol server that gives large language models browser automation through Playwright's accessibility tree, so agents can interact with web pages without screenshots or vision models.
Playwright MCP exposes browser automation as MCP tools backed by Playwright's structured accessibility snapshots rather than pixel-based screenshots, so it works with models that have no vision capability and avoids the ambiguity of screenshot-based interaction. The project notes that CLI- and skill-based workflows are often more token-efficient for coding agents doing routine browser tasks, since they avoid loading large tool schemas and accessibility trees into context, and suggests its own Playwright CLI with Skills for that case; it recommends the MCP server instead for specialized agentic loops that benefit from persistent state and iterative reasoning over page structure, such as exploratory automation or self-healing tests.
It requires Node.js 18 or newer and works with MCP clients including VS Code, Cursor, Windsurf, Claude Desktop, Claude Code, Codex, Goose and others, installed with a single JSON configuration block or a client-specific CLI command.
It is written in TypeScript and licensed under Apache-2.0.