A Model Context Protocol (MCP) server that gives AI coding assistants comprehensive browser automation and debugging capabilities, including execution-level inspection (logs, network, console messages) and visual feedback (screenshots, ARIA snapshots, PDFs). Version 0.6.9 is the current stable release; the package is actively developed on GitHub. Supports both Browser (Playwright) and Node.js (Inspector Protocol) platforms with session management, design comparison (Figma), OpenTelemetry integration, and multiple transport modes (stdio, HTTP). Differentiates from generic browser automation tools by focusing on AI agent integration via MCP, providing structured debugging tools (tracepoints, logpoints) and visual snapshots for LLM-based workflows.
npm install browser-devtools-mcpVerified import paths — ran on the pinned version, not inferred.
Shows programmatic instantiation of the Browser Platform and running the MCP server via runServer().
After tab closure, take a new snapshot (e.g., takeScreenshot) before using any element references.
Re-evaluate network state after tab recovery; avoid assumptions about pending requests.
Use --pid <PID> or --port <PORT> flags instead of --name.
Update calls to runServer({ platform, transport: 'stdio', sessionTimeout: 300000 }).npm install browser-devtools-mcp --save. Ensure project uses ESM (type: module in package.json) or use dynamic import if using CJS.
Ensure you import the platform class from the correct subpath, e.g., import { BrowserPlatform } from 'browser-devtools-mcp/platforms/browser'.For Node platform, start Node with --inspect flag. For browser platform, ensure Playwright browsers are installed: npx playwright install.