AIToday
Large Language ModelsAI Coding AssistantsOpen-Source AIZenn AI/MLPublished: Oct 27, 2026, 19:01 JST

chrome-use gives Claude Code logged-in Chrome, no debug port

Send on LINE
chrome-use gives Claude Code logged-in Chrome, no debug port

3 Key Points

  1. What happened

    chrome-use, an Apache-2.0 CLI and MCP server, connects agents such as Claude Code, Codex and Cursor to your everyday Chrome via a Web Store extension and native messaging, so no remote-debugging port is opened.

  2. Why it matters

    Agents inherit the tabs' existing login state, and each session runs in its own colored tab group, so a human can step in for two-factor authentication without touching other tabs.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The gap chrome-use targets is the login state agents leave behind. Most browser automation starts a fresh context with no cookies, so agents hit login screens, CAPTCHAs and new-device emails before doing any work. Playwright and Puppeteer can attach to an existing browser, but Chrome 136 and later refuse a --remote-debugging-port on the default profile, and Chrome 144's remote-debugging toggle makes every window in that profile visible to the tool and asks for permission on each connection.

chrome-use's answer is to skip the DevTools route entirely: a small extension talks to the CLI over native messaging and drives only the tabs the agent targets via chrome.debugger. The visible trade-off is the "debugging" bar Chrome shows while connected, which the --silent option removes by restarting Chrome with --silent-debugger-extension-api. Setup is a one-line install plus a store extension, and the project places a guidance skill for Claude Code, Codex and Cursor automatically.

The project positions itself against Chrome DevTools MCP and Playwright's extension mode by supporting multiple agents on the same logged-in Chrome without remote debugging; for front-end debugging or teams already on Playwright MCP, its own guidance points elsewhere. Its stealth measurements are self-reported and environment-specific, and the docs state plainly that no tool can promise CAPTCHAs never appear or that accounts stay safe.

FAQ
How does chrome-use connect to Chrome without a remote-debugging port?
It uses a Web Store extension plus native messaging, and the extension drives the target tab through chrome.debugger. Everything runs locally, with no network socket or remote server.
What happens when an agent hits a two-factor prompt or CAPTCHA?
The agent works in a visible window, so you can complete that step yourself in the same window and the agent continues. For Yidun slider puzzles it can also attempt solve-slider first.
Does chrome-use leave traces that bot-detection pages can see?
It injects no JavaScript patches and keeps Runtime.enable off by default, and it offers a --humanize mode. The project's own tests show 0% stealth and 0% headless on CreepJS, but the docs do not guarantee evasion or account safety.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleOpenAI AI disproves 指数時間仮説, proves 量子色力学 existence