π₯ Hot Repo: The MCP Server That Gives AI Agents Eyes in Your Browser
Google's official Chrome DevTools MCP server hit #1 trending on GitHub today with 315 stars gained in 24 hours. At 33,000+ total stars, it lets AI coding agents like Claude Code and Cursor debug live browser sessions, capture heap snapshots, and run Lighthouse audits β all via MCP.
By OMC Editorial on 2026-04-05
Chrome DevTools MCP: Google's Official Server Gives AI Coding Agents Full Browser Control
The ChromeDevTools/chrome-devtools-mcphttps://github.com/ChromeDevTools/chrome-devtools-mcp repository β Google's official Model Context Protocol server for Chrome DevTools β is the 1 trending repository on GitHub today, adding 315 stars in a single day to reach 33,246 total. Created in September 2025 by the Chrome DevTools team, the project has accumulated those stars in under seven months and is now the top browser-debugging integration listed in Claude Code, Gemini CLI, Cursor, and GitHub Copilot documentation.
What It Does
chrome-devtools-mcp turns Chrome DevTools into a set of MCP tools that any AI coding agent can call natively. Connect it to Claude Code with a single command:
Once connected, the agent can inspect live network traffic, read browser console logs with source-mapped stack traces, take full-page screenshots, run Lighthouse performance audits, and automate browser interactions via Puppeteer β all without leaving the coding loop. It officially supports Claude Code, Gemini CLI, Cursor, GitHub Copilot, Amp, and Antigravity.
The AutoConnect Feature: Agents Join Your Live Session
The most significant recent addition is --autoConnect, available since Chrome M144 Beta. Previously, agents launched their own isolated headless Chrome instance. With autoConnect, the agent attaches to the Chrome window you are already using, letting it work alongside you in a live session. This matters for debugging issues that require authentication, session state, or data that is difficult to recreate from scratch in a fresh browser.
There is a known tradeoff: in autoConnect mode, the server's PageCollector class can accumulate stale page entries without cleanup, causing memory growth of roughly 13Β MB/minute under heavy tab activity. Two active GitHub issues 1192, 1214 are tracking the regression.
v0.21.0: Memory Leak Detection for AI Workflows
Released AprilΒ 1, 2026, v0.