Oleksandr Kuzmenko, EditorPublished 8 min readAI-assisted · editor-reviewed
TL;DR
- Google released chrome-devtools-mcp as an official MCP server and CLI for Chrome automation in agentic IDEs.
- Supports console logging with source maps, network payload inspection, and CrUX-backed performance profiling.
- Includes pre-packaged agent skills for Largest Contentful Paint optimization, session debugging, and shell scripts.
- Offers a --slim mode to shrink the tool schema and save prompt tokens during agent interactions.
Key facts
- Supported Browsers
- Google Chrome, Chrome for Testing (Extended Stable)
- Telemetry Default
- Enabled (opt-out via --no-usage-statistics)
- Field Data Integration
- Google CrUX API (opt-out via --no-performance-crux)
- Interface Modes
- Model Context Protocol (MCP) server and chrome-devtools CLI
Direct DevTools Access for Agentic IDEs
Google's chrome-devtools-mcp connects coding agents like Claude Code, Cursor, VS Code Copilot, and Gemini CLI directly to Google Chrome or Chrome for Testing. Rather than relying on rigid browser automation scripts, assistants gain direct programmatic control over Chrome DevTools APIs to inspect DOM trees, capture screenshots, audit network traffic, and read runtime console errors with mapped source locations.
The server starts the browser lazily upon the first tool call that requires an active session. Engineers can also invoke the included chrome-devtools CLI independently in terminal scripts:
chrome-devtools navigate_page 1 --url "https://developers.chrome.com" chrome-devtools take_screenshot 1 --filePath screenshot.png
Agent Skills and Performance Audits
The repository bundles modular agent skills that teach LLMs specific debugging workflows, including inspecting cookies, debugging authentication state, scripting the CLI, and resolving Largest Contentful Paint (LCP) issues. In Claude Code, the package installs as a plugin via /plugin marketplace add ChromeDevTools/chrome-devtools-mcp followed by /plugin install chrome-devtools-mcp@chrome-devtools-plugins.
Performance audits run through DevTools can automatically query the Chrome User Experience Report (CrUX) API to overlay real-user field data over synthetic lab traces, giving agents holistic context when fixing performance regressions.
Privacy Flags and Context Footprint
By default, usage telemetry and CrUX queries are active. Teams operating in private environments can disable data collection by passing --no-usage-statistics or defining the CHROME_DEVTOOLS_MCP_NO_USAGE_STATISTICS environment variable, and turn off CrUX calls with --no-performance-crux. Update notifications from npm can be silenced via CHROME_DEVTOOLS_MCP_NO_UPDATE_CHECKS.
To prevent the MCP tool catalog from exhausting context windows during iterative reasoning loops, passing the --slim flag restricts the registered tool surface to basic navigation and inspection endpoints.
Try it in 2 minutes
{
"mcpServers": {
"chrome-devtools": {
"command": "npx",
"args": [
"-y",
"chrome-devtools-mcp@latest",
"--no-usage-statistics",
"--slim"
]
}
}
}When to use
- Debugging frontend application state and console errors directly from Claude Code, Cursor, or Copilot.
- Auditing Core Web Vitals and resolving Largest Contentful Paint bottlenecks with agent assistance.
- Executing authenticated multi-step user flow tests against live Chrome instances.
When NOT to use
- Headless production web scraping pipelines that do not involve LLM reasoning loops.
- Non-Chromium browser automation tasks targeting Firefox or Safari.
- Strictly offline air-gapped environments without disabling external CrUX and npm update network checks.
What to do today
- Add chrome-devtools-mcp to your MCP client configuration using the --no-usage-statistics flag if privacy is required.
- Use the --slim flag in token-constrained setups to prevent large tool definitions from inflating agent prompt costs.
- Test automated UI debugging by prompting your agent to inspect console errors or performance traces on a local dev server.
Step names are taken from What to do today.
Step names are taken from What to do today. Add chrome-devtools-mcp to your MCP client configuration using the --no-usage-statistics flag if privacy is required.. Use the --slim flag in token-constrained setups to prevent large tool definitions from inflating agent prompt costs.. Test automated UI debugging by prompting your agent to inspect console errors or performance traces on a local dev server.
- Add chrome-devtools-mcp to your MCP client configuration using the --no-usage-statistics flag if privacy is required.
- Use the --slim flag in token-constrained setups to prevent large tool definitions from inflating agent prompt costs.
- Test automated UI debugging by prompting your agent to inspect console errors or performance traces on a local dev server.
