home / mcp / chrome devtools mcp server
Provides a local MCP server to control Chrome DevTools for automation, debugging, and performance analysis.
Configuration
View docs{
"mcpServers": {
"chromedevtools-chrome-devtools-mcp": {
"command": "npx",
"args": [
"-y",
"chrome-devtools-mcp@latest"
],
"env": {
"CHROME_DEVTOOLS_MCP_NO_USAGE_STATISTICS": "(unset)"
}
}
}
}You can run a local Chrome DevTools MCP server that lets your AI assistants control and inspect a live Chrome browser. This server provides reliable automation, advanced debugging, and performance analysis capabilities through a Model-Context-Protocol interface.
Install and run the MCP server locally, then connect your MCP client to it. Use the standard or slim mode to tailor capabilities. With the server active, you can perform browser automation, collect performance traces, inspect network activity, capture screenshots, and read console output. The server will automatically start a Chrome instance when needed, and you can also connect to an already running Chrome for shared state.
Prerequisites: you need Node.js (version 16+ or newer as appropriate for your environment), npm, and a compatible Chrome browser. Use the following concrete steps to install and run the Chrome DevTools MCP server.
To use the Chrome DevTools MCP server with your MCP client, configure the client to point at the local MCP server using the provided npm-based command. The standard setup runs the server with automatic installation and the latest release.
Be aware that the MCP server exposes the browser contents to MCP clients. Avoid sharing sensitive or personal information through the MCP channel. Performance traces may query the CrUX API for field data unless you disable that behavior.
If you encounter issues starting the browser or connecting, ensure remote debugging is not blocked in your environment, and verify that the browser instance is accessible according to your chosen connection method.
Run a simple prompt to verify that you can check a page’s performance and get a trace recorded by the MCP client. The client should open a browser, load the target page, and produce a trace.
A group of tools to interact with page elements by clicking, dragging, typing, filling fields, handling dialogs, and uploading files.
Tools to navigate pages, manage multiple pages, wait for conditions, and switch between pages.
Tools to emulate devices or viewport sizes to test responsive behavior.
Tools to start and stop performance traces, analyze insights, and capture memory snapshots.
Tools to fetch and inspect network requests to diagnose loading issues.
Tools to evaluate scripts, read console messages, run audits, and take visual snapshots.