microsoft/playwright-mcp

What it solves

Playwright MCP allows Large Language Models (LLMs) to interact with web pages without needing vision models or screenshots. It solves the problem of ambiguity and high token costs associated with pixel-based browser automation by providing LLMs with structured accessibility snapshots of web pages.

How it works

It acts as a Model Context Protocol (MCP) server that interfaces with Playwright. Instead of sending images of a page, it provides the LLM with a structured accessibility tree, allowing the model to reason over the page structure and perform actions deterministically.

Who it’s for

Developers and AI agent creators building specialized agentic loops that require persistent browser state, rich introspection, and iterative reasoning over web page structures, such as for exploratory automation or self-healing tests.

Highlights

• Accessibility-based: Uses structured data rather than pixels, removing the requirement for vision models.
• Deterministic: Avoids the ambiguity common in screenshot-based browser automation.
• Lightweight: Fast and token-efficient compared to visual inputs.
• Broad Client Support: Compatible with a wide range of MCP clients including VS Code, Cursor, Claude Desktop, and Windsurf.