Playwright MCP

Playwright MCP is Microsoft's MCP server for the Playwright browser automation framework. Its README describes the approach: the server "enables LLMs to interact with web pages through structured accessibility snapshots, bypassing the need for screenshots or visually-tuned models." Tools cover navigation, input, and inspection of a live browser; one tool, browser_run_code_unsafe, is labeled by the README itself as executing arbitrary JavaScript in the Playwright server process.

The documented way to run it is local: npx @playwright/mcp@latest over stdio, an optional self-hosted HTTP mode behind the --port flag, or the mcr.microsoft.com/playwright/mcp Docker image. Microsoft hosts no public remote endpoint for it. There is no MCP-level authentication; browser sessions carry state through Playwright's storage-state mechanism, and the README states plainly that the server "is not a security boundary."

No Microsoft surface names a supported MCP specification revision. The repository has no changelog file, release notes through v0.0.79 (2026-08-06) describe only tool changes, and the package declares @modelcontextprotocol/sdk ^1.25.2 as a devDependency, which is not evidence of revision support. spec_status is therefore unknown.

The README also steers part of its audience elsewhere: it recommends coding agents use the companion microsoft/playwright-cli project with skills instead of MCP "because CLI invocations are more token-efficient," while keeping MCP for exploratory automation and long-running workflows. The server remains under active development, with releases in August 2026.