Browser DevTools MCP
Enables comparison of web pages against Figma designs for visual consistency.
Allows automation and debugging of Firefox browser, including navigation, screenshots, and network inspection.
Allows automation and debugging of Google Chrome browser, including navigation, screenshots, and network inspection.
Integrates distributed tracing with trace context propagation for monitoring and debugging.
Provides React DevTools for inspecting React components and elements within the browser.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Browser DevTools MCPNavigate to google.com and take a full-page screenshot"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Browser DevTools MCP for VS Code & Cursor
Playwright-powered browser automation and debugging for VS Code and Cursor via the Model Context Protocol (MCP).
This extension integrates browser-devtools-mcp into your IDE, enabling AI assistants like GitHub Copilot and Cursor AI to interact with real web browsers for testing, debugging, and automation tasks.
Features
🌐 Browser Automation - Navigate, click, fill forms, and interact with web pages
📸 Screenshots - Capture full-page or element screenshots
♿ Accessibility - Run accessibility audits and get ARIA/AX tree snapshots
📊 Web Vitals - Measure Core Web Vitals (LCP, INP, CLS, TTFB, FCP)
🔍 Network Inspection - Monitor HTTP requests and responses
🎭 Request Mocking - Stub and mock API responses
⚛️ React DevTools - Inspect React components and elements
🔭 OpenTelemetry - Distributed tracing integration with trace context propagation
🎨 Figma Comparison - Compare pages with Figma designs
🐛 Non-Blocking Debugging - Tracepoints, logpoints, exceptionpoints, watch expressions, probe snapshots
⚡ Execute - Batch multiple tool calls in one request via JavaScript and
callTool(); on browser platformpage(Playwright Page) is available forpage.evaluate(),page.locator(), etc.🌐 Playwright Browsers - On first install/upgrade, the extension downloads the browsers selected in settings (default: Chromium + headless shell + ffmpeg) into Playwright’s normal cache using
playwright-core’s installer—nonpxrequired. VSIX builds setPLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1so binaries are not bundled. If that step fails for the Chromium stack, you may be prompted to use installed Google Chrome instead (Use system-installed browser). Run Browser DevTools MCP: Install Playwright Browsers... anytime to pick Chromium (default), Firefox, and/or WebKit: it updatesinstall.chromium/install.firefox/install.webkitto match and downloads those engines.📦 MCP Server - Shipped inside one universal VSIX (same artifact for all platforms). The package includes
sharp+@img/sharp-wasm32; native sharp/libvips prebuild variants are excluded from VSIX contents. No npm required at runtime; the extension runs the bundledbrowser-devtools-mcpwithnode.
Related MCP server: Playwright MCP
Installation
From Open VSX Registry
Open VS Code or Cursor
Go to Extensions (
Ctrl+Shift+X/Cmd+Shift+X)Search for "Browser DevTools MCP"
Click Install
Or install via command line:
# VS Code
code --install-extension serkan-ozal.browser-devtools-mcp-vscode
# Cursor
cursor --install-extension serkan-ozal.browser-devtools-mcp-vscodeFrom VSIX
# VS Code
code --install-extension browser-devtools-mcp-vscode-x.x.x.vsix
# Cursor
cursor --install-extension browser-devtools-mcp-vscode-x.x.x.vsixRegistration: In Cursor the extension registers the MCP server via Cursor’s native MCP API (no mcp.json needed). In VS Code 1.96+ it uses vscode.lm.registerMcpServerDefinitionProvider. The server is started automatically when the extension is enabled.
Telemetry
The extension can send anonymous usage events (install/uninstall, browser install step, etc.) to help improve the product. The same opt-in/opt-out rules apply as for the bundled browser-devtools-mcp server. No PII is collected; only an anonymous ID in ~/.browser-devtools-mcp/config.json, plus event name and environment properties (e.g. extension version, OS, Node version).
Events:
cursor_ext_activated/cursor_ext_deactivated(extension lifecycle),cursor_ext_installed(only when first-install/upgrade path ran and bundled MCP path resolved),cursor_ext_install_failed,cursor_ext_browser_installed/cursor_ext_browser_install_failed, andcursor_ext_uninstalled(when the extension is uninstalled and deactivate runs with the extension listed in.obsolete). MCP registration state is included where relevant (mcp_server_registered/mcp_server_unregistered). If telemetry is disabled (setting, env, or config), no events are sent.Timing: Clients may batch or delay sending—events might not appear in analytics immediately.
TELEMETRY_ENABLE=falseand~/.browser-devtools-mcp/config.jsonapply to both the extension and the bundled MCP process (the extension forwards the parent environment to the server).
How to disable telemetry
Setting (recommended): Set
browserDevtoolsMcp.telemetry.enabletofalsein VS Code/Cursor settings. The extension syncs this to~/.browser-devtools-mcp/config.jsonon activate and when the setting changes, so no telemetry events (including uninstall) are sent.Environment variable: Set
TELEMETRY_ENABLE=falsebefore starting VS Code/Cursor.Config file: Edit
~/.browser-devtools-mcp/config.jsonand set"telemetryEnabled": false.
MCP Server (bundled)
The browser-devtools-mcp package is a dependency of this extension and is included in the published VSIX with production node_modules. We publish a single universal VSIX where native sharp/libvips prebuild variants are excluded from the bundle and @img/sharp-wasm32 is included. @img/sharp-wasm32 is a direct extension dependency (aligned with the bundled sharp), and .npmrc sets force=true so npm installs it on normal hosts.
Activate: The extension resolves
node_modules/browser-devtools-mcp/dist/index.jsinside the extension folder. No npm and no network required for the server binary itself.New server versions: Publish/build workflows are lockfile-driven (
npm ci --omit=optional). Bumpbrowser-devtools-mcpinpackage.jsonand refresh lockfile when you want to roll forward.Dependencies:
browser-devtools-mcpand@img/sharp-wasm32are regular dependencies inpackage.jsonfor maintainers; end users do not run npm for MCP.
Maintainer: universal VSIX
PR / CI: .github/workflows/build.yml —
workflow_dispatch,pull_request(master), andpush(main) triggers; onubuntu-latestit runsnpm ci --omit=optional, lint, build, andnpx vsce package.Release / Open VSX: .github/workflows/publish-vscode-extension.yml — single
releasejob runsnpm ci --omit=optional, lint, build, version bump/tag/release, then publishes to Open VSX via HaaLeo/publish-vscode-extension@v2 (skipDuplicate: true) and uploads the produced VSIX as artifact.Packaging filter:
.vscodeignoreexcludes@img/sharp-darwin-*,@img/sharp-win32-*,@img/sharp-linux-*,@img/sharp-libvips-*and keepssharp+@img/sharp-wasm32.vscecollects production dependencies vianpm list --production, so devDependencies are not bundled.
Local packaging check:
npm ci --omit=optional
npx vsce packagePlaywright Browsers
The extension uses Playwright’s browser binaries (Chromium, Firefox, WebKit), stored in the default cache (e.g. ~/.cache/ms-playwright on Linux, ~/Library/Caches/ms-playwright on macOS).
Which browsers to install: In Settings, use
browserDevtoolsMcp.install.chromium,install.firefox, andinstall.webkit(default: Chromium group). On first install/upgrade, the extension calls Playwright’sinstallBrowsersForNpmInstallwith that selection (unless Use system-installed browser is on or platform is Node). Playwright skips work when the chosen builds are already present in the cache (INSTALLATION_COMPLETE). Changing these settings triggers another install pass; restart the MCP session if it was already running. Install Playwright Browsers... in the Command Palette does the same selection UI and writes those three settings to match your choice, then runs the installer (so the sidebar/settings panel stay in sync).VSIX / CI:
PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1is set when packaging so browser ZIPs are not part of the extension.Download failures (Chromium stack): If Playwright’s install step fails while the Chromium group is requested (network, proxy, firewall, Winldd on Windows, etc.), the extension shows one notification (error text in details) and may offer Use Google Chrome to enable Use system-installed browser (
browserDevtoolsMcp.browser.useSystemBrowser). Google Chrome must be installed on the machine. Firefox/WebKit-only installs get a generic failure message instead. Restart the MCP session (or Restart Server) after accepting. You can also turn on Use system-installed browser manually in settings anytime.
To skip browser download (e.g. you use a system browser or custom path), set the environment variable PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1 before starting VS Code/Cursor.
Configuration
Quick Settings
Open the Browser DevTools MCP panel in the Explorer sidebar to configure common settings.
Full Settings
Open VS Code Settings (Ctrl+, / Cmd+,) and search for "Browser DevTools MCP" or use the command:
Browser DevTools MCP: Open SettingsTo (re)download Playwright binaries immediately:
Browser DevTools MCP: Install Playwright Browsers...Available Settings
Settings below are passed to the MCP server as environment variables. Change them in Settings or in the Browser DevTools MCP sidebar panel (subset); restart the MCP session to apply.
General
Setting | Default | Description |
|
| Enable or disable the extension (and MCP server) |
|
| Allow anonymous install/uninstall telemetry (see Telemetry) |
|
| MCP platform: |
Install (Playwright browsers to download on activate)
Setting | Default | Description |
|
| Browser group selection used on first install/upgrade; changing it triggers Playwright install pass. |
|
| Browser group selection used on first install/upgrade; changing it triggers Playwright install pass. |
|
| Browser group selection used on first install/upgrade; changing it triggers Playwright install pass. |
Browser
Setting | Default | Description |
|
| Run browser in headless mode |
|
| Enable persistent browser context |
|
| Directory for persistent user data |
|
| Use system browser instead of bundled |
|
| Custom browser executable path |
|
| Browser locale (e.g., en-US, tr-TR) |
|
| Enable CDP attach mode (Chromium only). |
|
| Optional CDP endpoint ( |
|
| On loopback CDP failure, open |
|
| Max console messages to buffer |
|
| Max HTTP requests to buffer |
Node (when platform is node)
Setting | Default | Description |
|
| Inspector host for Docker (e.g., host.docker.internal) |
|
| Max console messages to buffer in Node process |
OpenTelemetry
Setting | Default | Description |
|
| Enable OpenTelemetry instrumentation |
|
| Service name for traces |
|
| Service version for traces |
|
| OpenTelemetry assets directory |
|
| Comma-separated events to instrument (default: click) |
|
| Exporter: |
|
| OTLP collector URL |
|
| HTTP headers for collector |
AWS / Amazon Bedrock
Setting | Default | Description |
|
| AWS region for Bedrock |
|
| AWS profile name |
|
| Enable Bedrock for AI features |
|
| Image embedding model ID |
|
| Text embedding model ID |
|
| Vision model ID |
Figma
Setting | Default | Description |
|
| Figma API access token |
|
| Figma API base URL (default: https://api.figma.com/v1) |
Advanced (MCP)
Setting | Default | Description |
|
| Omit tool output schema from MCP registration (can reduce token usage) |
|
| Comma-separated domains to enable (e.g. navigation,interaction,a11y). Empty = all. Browser: a11y, content, debug, figma, interaction, navigation, o11y, react, run, stub, sync. Node: debug, run. |
Usage
Once installed, the MCP server is automatically available to AI assistants. Try prompts like:
Navigation & Screenshots:
Navigate to https://example.com and take a screenshot
Take a full-page screenshot of the current page
Wait for network to be idle and then take a screenshotAccessibility Testing:
Check the accessibility of the current page
Get the ARIA snapshot for the navigation menu
Get the AX tree snapshot with occlusion checking enabledPerformance:
Get the Web Vitals for https://google.com
What is the LCP score of this page?
Measure Core Web Vitals and give me recommendationsInteraction:
Fill the login form with test@example.com and click submit
Click the "Sign Up" button and wait for the page to load
Scroll to the bottom of the pageDebugging:
Show me the console errors on this page
What network requests failed on this page?
Set a tracepoint at line 50 in main.js and capture the call stack
Get probe snapshots after triggering the code pathAPI Mocking:
Mock the /api/users endpoint to return an empty array
Intercept all API requests and add an auth header
List all active stubs and clear themExecute (batch tool calls + optional page script):
Use execute to fill the login form and click submit in one call
Run a script that calls callTool('navigation_go-to', { url: '...' }) then callTool('a11y_take-aria-snapshot', {}, true)Visualizer UI
Visualizer UI screenshot:
![]()
Visualizer UI renders MCP tool events as an animated scene. The UI connects to the Visualizer WebSocket stream at ws://localhost:<wsPort>.
How to open
Open the
Browser DevTools MCPpanel in VS Code/Cursor.Set
Show Visualizertotrue.Run
Browser DevTools MCP: Restart Server.Run any MCP tool (for example
navigation_go-to, thencontent_take-screenshot).
If the panel does not open automatically, open Command Palette (Cmd+Shift+P) and run Browser DevTools MCP: Show Visualizer.
Alternative (run Visualizer UI locally for development):
From repo root:
cd visualizer-ui
npm install
npm run dev
Open http://localhost:3000 in your browser.
How it works (summary)
The extension starts a Visualizer WebSocket server on
browserDevtoolsMcp.visualizer.wsPort.The UI reads
VIS_WS_PORTinvisualizer-ui/src/main.tsand connects tows://localhost:<wsPort>.Tool events (
run_started,tool_started,tool_finished,run_done, etc.) are represented in the scene.
Settings you should configure
Show Visualizer: set totrueto enable both the webview UI and the Visualizer WebSocket server.browserDevtoolsMcp.visualizer.wsPort(default3020): change this if the port is busy, then runRestart Server.
Available MCP Tools
Navigation Tools
Tool | Description |
| Navigate to a URL |
| Reload the page |
| Go back or forward in history (direction: back | forward) |
Content Tools
Tool | Description |
| Take a screenshot (full page or element) |
| Get page HTML with filtering options |
| Get visible text content |
| Save page as PDF |
Interaction Tools
Tool | Description |
| Click an element |
| Fill an input field |
| Hover over an element |
| Scroll page or element |
| Press a keyboard key |
| Drag and drop |
| Select dropdown option |
| Resize viewport (Playwright emulation) |
| Resize browser window (OS-level) |
Accessibility Tools
Tool | Description |
| Get ARIA snapshot (YAML format) |
| Get AX tree with visual diagnostics |
Observability Tools
Tool | Description |
| Get Web Vitals (LCP, INP, CLS, TTFB, FCP) |
| Get console logs with filtering |
| Get network requests with filtering |
| Get current OpenTelemetry trace ID |
| Generate new trace ID |
| Set trace ID for distributed tracing |
Synchronization Tools
Tool | Description |
| Wait for network to become idle |
Stub Tools
Tool | Description |
| Mock HTTP response |
| Intercept and modify requests |
| List installed stubs |
| Clear stubs |
React Tools
Tool | Description |
| Get React component for DOM element |
| Get DOM element for React component |
Execute
Tool | Description |
| Batch-execute multiple tool calls in one request via JavaScript; use |
Figma Tools
Tool | Description |
| Compare page with Figma design |
Debug Tools (Non-Blocking)
Tool | Description |
| Set a tracepoint (captures call stack) |
| Set a logpoint (evaluates expression) |
| Configure exception breakpoints (none, uncaught, all) |
| Add watch expression (evaluated at every tracepoint hit) |
| Remove a tracepoint, logpoint, or watch by type and id |
| List tracepoints, logpoints, and/or watches |
| Clear tracepoints, logpoints, and/or watch expressions |
| Get tracepoint, logpoint, and/or exceptionpoint snapshots |
| Clear probe snapshots (optional types, probeId) |
| Get debugging status (probe counts, exceptionpoint state) |
| Resolve bundle location to original source via source maps |
When using Node platform (browserDevtoolsMcp.platform: node), additional tools: debug_connect, debug_disconnect, debug_get-logs.
Development
Prerequisites
Node.js 22+
VS Code 1.96+ or Cursor
Build
# Install dependencies
npm install
# Compile TypeScript
npm run compile
# Watch for changes
npm run watch
# Package extension
npm run packageTesting
Press
F5in VS Code/Cursor to launch Extension Development HostThe extension will be loaded in the new window
Open AI Chat (Copilot or Cursor AI) and test MCP tools
Troubleshooting
Restart the MCP server
If you run into problems—for example the MCP server fails to start, the browser that was opened has closed, MCP processes have leaked, or you see other odd behavior—try restarting the MCP server first:
Open the Command Palette (
Ctrl+Shift+P/Cmd+Shift+P)Run Browser DevTools MCP: Restart Server
This unregisters the server, stops any running MCP processes (e.g. Cursor-started ones), and registers it again so a fresh process is started.
MCP Server Not Starting
Reinstall the extension or install a fresh VSIX—the MCP server is bundled; a missing
dist/index.jsusually means a broken or partial install.Check that Node.js 22+ is available to the IDE (extension host) and the extension is enabled.
Check Output panel for "Browser DevTools MCP" logs.
Browser Not Launching
Run Browser DevTools MCP: Install Playwright Browsers... and ensure Chromium (or your engine) is selected, or rely on
install.*settings on first install/upgrade. If the Playwright download fails, enable Use system-installed browser in settings or choose Use Google Chrome when the extension prompts (Chromium stack only; Chrome must be installed). Skip download if you use a system browser only.Try disabling headless mode in settings
Check if a custom executable path is needed (e.g. system browser or custom build)
Settings Not Applying
After changing settings, restart the MCP session:
Run the command Browser DevTools MCP: Restart Server
Or reload the VS Code/Cursor window (
Cmd+Shift+P→ "Developer: Reload Window")
Contributing
Contributions are welcome! Please open an issue or submit a pull request on GitHub.
License
MIT License - see LICENSE for details.
Links
GitHub Repository - Extension source code
Open VSX Registry - Extension page
browser-devtools-mcp - Main MCP server (npm)
This server cannot be installed
Maintenance
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/serkan-ozal/browser-devtools-mcp-vscode-extension'
If you have feedback or need assistance with the MCP directory API, please join our Discord server