OmniscioGuidesClaude Code and Chrome
Claude Code and Chrome: let it drive the browser
Chrome integration closes the loop that usually costs a dozen terminal switches: Claude writes the change, then opens the page, clicks through it and reads the console for what broke.
This page covers what it can do, how to connect it, what decides which sites it may touch, and the failures people actually hit.
Last checked against Anthropic's Claude Code documentation on November 18, 2026. By the Omniscio team. Updated .
What it can do
With Chrome connected, browser actions and coding work sit in one conversation. Anthropic's list of what that buys you:
- Live debugging — read console errors and DOM state, then fix the code that caused them.
- Web app testing — form validation, visual regressions, user flows.
- Design verification — build a UI from a mock, then open it and compare.
- Authenticated apps — Google Docs, Gmail, Notion, anything you are already signed into, with no API connector.
- Data extraction and task automation — pull structured information off a page, or fill forms and run multi-site workflows.
- File uploads and session recording — attach files from your machine, and record interactions as a GIF.
Two behaviours shape everything else. Claude opens new tabs for its work rather than taking over yours, and it shares your browser's login state, which is why an authenticated app needs no connector. Browser actions run in a visible window in real time, and when Claude meets a login page or a CAPTCHA it pauses and asks you to handle it.
Sources: the intro and "Capabilities" in Anthropic's Chrome page, read November 18, 2026.
What it needs
- A Chromium browser: Chrome or Edge are the supported pair, and Claude Code also detects Brave, Arc, Vivaldi and Opera.
- The Claude in Chrome extension, version 1.0.36 or later, from the Chrome Web Store.
- Claude Code, and a direct Anthropic plan — Pro, Max, Ultra, Team or Enterprise.
- A claude.ai sign-in. This one excludes people quietly: with an API key or a long-lived token from
claude setup-token, Claude Code keeps Chrome integration off even when you pass--chrome, because the extension cannot authenticate with those credentials.
It is also not available through Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. If you reach Claude only through one of those, this feature needs a separate claude.ai account.
Two policies can switch it off. In an Enterprise organization with HIPAA enabled it is off by default until an Owner turns it on. And with Zero Data Retention enabled for Claude Code it is unavailable in Claude Code entirely: /chrome reports that your organization's policy disables it.
Sources: "Prerequisites" and "Chrome availability with HIPAA or zero data retention" in Anthropic's Chrome page, read November 18, 2026.
Connect it: --chrome and /chrome
Launch with the flag:
claude --chromeThe first launch shows a one-time dialog explaining the integration and how site permissions work. Press Enter and the session has browser tools.
Then ask for the thing, not the tool:
Go to code.claude.com/docs, click on the search box,
type "hooks", and tell me what results appear/chrome is the status panel and the control room. It reports whether the integration is on — the working state reads Status: Enabled and Extension: Installed — and from it you reconnect the extension or, when more than one browser is connected, choose which one Claude uses. With several connected, a browser action that starts before you have chosen prompts you to pick, and Claude keeps your choice even when another browser connects later.
If the extension is missing in an interactive session, Claude Code offers to install it once per session, with three answers: install it, not now, or do not ask again. Two managed MCP policies suppress that prompt entirely — a deniedMcpServers rule that blocks the claude-in-chrome server, or a managed-mcp.json that does not allow it alongside the managed set.
Sources: "Get started in the CLI" and "Install the extension when Claude asks" in Anthropic's Chrome page, read November 18, 2026.
Permissions and the tab group
Site-level permissions come from the extension, not from Claude Code: manage them in the extension's settings to control which sites Claude can browse, click and type on. When Claude asks before an action, the dialog starts with Claude in Chrome wants to and offers to allow all actions on that site for the session.
The tabs Claude opens are collected into a Chrome tab group tied to the session, and whether that group closes when the session ends depends on how it ended. On /clear, Claude Code closes the group, open pages included, unless work that survives the clear is still running. Switching sessions with /resume, exiting, or clearing while such work runs closes the group only if it holds nothing but empty new tabs — so pages you might still be reading stay open.
In a VS Code session the permission prompt arrives as a card in the chat panel instead, and for a site you have not allowed the card also offers to allow it.
Sources: "Manage site permissions", the tab-group paragraph in the intro, and "Permission prompts in VS Code sessions" in Anthropic's Chrome page, read November 18, 2026.
Enabled by default, and what it costs
So you do not type the flag every time, run /chrome in a CLI session and select Enabled by default. The VS Code extension has the same switch and shares the same setting, so turning it on in one turns it on for both.
The price is context. Anthropic says enabling it by default increases context usage, because the browser tools and their instructions are always loaded. If you notice that, turn it off and connect only when a task needs it.
The startup dialog is the same setting under another name: Yes, use my browser turns it on and connects the session; No, keep browser tools off, or Esc, turns it off and stops Claude Code offering it again.
Sources: "Enable Chrome by default", its note, and "Answer the startup dialog" in Anthropic's Chrome page, read November 18, 2026.
Why a project cannot turn it on
A repository's .claude/settings.json can set environment variables, so it can try to set CLAUDE_CODE_ENABLE_CFC to 1 and switch Chrome on. Claude Code ignores it and prints a warning saying so.
Those files live in the project directory, and a repository you check out must not be able to connect Claude to your browser. Turning it on is your decision, made with --chrome or /chrome, in a session you started.
So there are two fixes and they are different. For browser tools now, exit and restart with claude --chrome. For them in later sessions, run /chrome and select Enabled by default — which applies to sessions you start afterward, not the one you are in.
Source: "Project settings can't turn on Chrome" in Anthropic's Chrome page, read November 18, 2026.
When it stops working
Extension not detected. Check that it is installed and enabled at chrome://extensions, that Claude Code is current, and that Chrome is running. Then run /chrome and select Reconnect extension. The first time you enable the feature, Claude Code writes a native messaging host configuration file that Chrome reads at startup — so if it failed on the very first attempt, restart Chrome.
The browser stops responding. Look for a modal dialog on the page first: JavaScript alerts block browser events and Claude stops receiving commands. Dismiss it and tell Claude to continue. Otherwise ask Claude for a new tab, or disable and re-enable the extension.
It drops during a long session. The extension's service worker can go idle, which breaks the connection. Run /chrome, read the Status line, and select Reconnect extension. Before v2.1.290 that menu item only opened a page and did not restart a failed connection, so on an older version update Claude Code.
"Browser extension is not connected" when you are signed in. If you belong to more than one claude.ai organization, the extension must be signed in to the same one as Claude Code. Run /status and read the Organization row, then log out in the extension and back in to that organization. Anthropic warns that logging out of the extension loses the shortcuts and scheduled tasks saved in it, so try the other fixes first.
On Windows there are three of its own: a named pipe conflict (EADDRINUSE) that a restart and closing other sessions clears, a native messaging host that crashes on startup and wants a reinstall, and setup pages that fail to open before v2.1.211.
Sources: "Troubleshooting", "Windows-specific issues" and "Common error messages" in Anthropic's Chrome page, read November 18, 2026.
What it is not
- Not a WSL feature. Anthropic states plainly that it is not supported under Windows Subsystem for Linux.
- Not for a zero-retention organization. It stays off there even if an Owner turned it on in the admin settings.
- Not a headless browser. Actions run in a visible window, because you may have to step in for a login or a CAPTCHA.
- Not native app control. Native macOS applications are a different feature, computer use; Chrome integration itself is the same in the CLI and in the VS Code extension.
Where these answers come from
- Everything on this page: Use Claude Code with Chrome, read November 18, 2026.
- The permission modes the prompts sit inside: Claude Code permission modes.
- Working inside the editor rather than a terminal: Claude Code in VS Code.
- The sandbox that browser tools are not part of: Claude Code sandbox.
Common questions
Why does /chrome say the extension is not detected?
The first time Claude Code enables the integration it writes a native messaging host configuration file, and Chrome reads that file at startup. So if it failed on the very first attempt, restart Chrome, then run /chrome and select Reconnect extension.
Why can a project setting not turn it on?
Claude Code ignores CLAUDE_CODE_ENABLE_CFC in a project's settings file on purpose: a repository you check out must not be able to connect Claude to your browser.
Does it work with an API key or Bedrock?
No. The extension authenticates with a claude.ai sign-in, and the feature is unavailable through Bedrock, Google Cloud's Agent Platform and Microsoft Foundry.
Does it work in WSL, or on Windows?
Not in WSL. On Windows it works, with its own failure modes: named pipe conflicts, a host that crashes on startup, and setup pages that fail to open before v2.1.211.