MCP Server
Playwright MCP
Microsoft's browser automation MCP server, using Playwright and structured accessibility snapshots for deterministic page interaction.
What is Playwright MCP?
Playwright MCP gives an agent a persistent browser session and structured access to page semantics. Its default interaction model relies on accessibility snapshots rather than screenshots, allowing the model to inspect element roles and invoke deterministic browser tools without a vision model. It is especially well suited to exploratory automation, self-healing test investigation, and longer loops that benefit from retaining browser state. The project also documents the tradeoff with Playwright CLI plus Skills, which can be more token-efficient for coding agents that do not need a large persistent browser tool surface.
Key capabilities
Semantic page state
Uses the accessibility tree so agents reason over roles and element references instead of pixel coordinates.
Persistent browser context
Supports iterative exploration where cookies, navigation state, and open pages matter across several tool calls.
Extensive runtime controls
Offers browser choice, device emulation, network origin controls, permissions, headless mode, and connection options.
Use cases
Explore and test a web application
Inspect live page structure, reproduce user flows, and collect deterministic evidence during debugging.
MCP capabilities
Tools
Page navigation
Open pages and maintain a browser session across an agent workflow.
Browser interaction
Click, type, select, inspect console output, and perform other deterministic browser actions.
Resources
Accessibility snapshots
Expose structured page state without requiring screenshot interpretation.
Connection and auth
- Node.js 18 or newer
- An MCP client
- A browser available locally or through the provided container
Client configuration
Generic MCP client
{"mcpServers":{"playwright":{"command":"npx","args":["@playwright/mcp@latest"]}}}