Clear, practical technology insights BSOD Code Lookup · Windows Error Code Lookup · Wi-Fi Troubleshooting · PC Troubleshooting Checklist

How to Use OpenAI Codex with Google Chrome

Connect the official Chrome extension to ChatGPT Work or Codex, choose between Chrome and the built-in browser, start tasks, and manage permissions safely.

Table of Contents

The Codex Chrome extension lets ChatGPT Work or Codex use your existing Chrome profile, including open tabs and sites where you are already signed in. It is useful for browser tasks that need your real session, while the built-in browser is better for public pages, localhost, or work you want to keep separate from your Chrome profile.

How to use OpenAI Codex directly in Google Chrome Picture 1

What the Chrome extension can do

With permission, ChatGPT can read a page, navigate, fill forms, work across tabs, and use web applications such as a CRM or analytics dashboard. Tabs created for a task are grouped so they are easier to review. Because this runs in your Chrome context, the active website account and its permissions matter.

Set up Chrome access

  1. Open the ChatGPT desktop app on a supported Windows or macOS computer.
  2. Switch to Work or Codex.
  3. Open Plugins > Computer Use and install or enable the plugin if prompted.
  4. In the Computer Use settings, enable the required server and skill controls, then connect Google Chrome.
  5. Install the prompted Chrome extension from the official Chrome Web Store listing. Avoid similarly named third-party extensions.
  6. Confirm that Chrome shows as connected before starting a task.

Availability depends on the ChatGPT plan, workspace settings, operating system, and region. If a control in this guide is absent, check the official Computer Use setup page rather than installing an unofficial substitute.

Start a task from Chrome

  1. Open the page you want to work with.
  2. Select ChatGPT from the Chrome toolbar or Extensions menu.
  3. Ask a question about the page or describe a bounded action.

On macOS, the documented shortcut is Command + Shift + . when the feature is available. The extension can use context from open tabs when the task requires it.

Start a Chrome task from ChatGPT

Begin a new Work or Codex chat and describe the result you want. ChatGPT can select Chrome when the task needs a signed-in website. You can also request it explicitly:

@Chrome open Salesforce and draft an update from these call notes. Do not save until I review it.

A scoped prompt is safer than “manage my Salesforce.” Name the website, record, allowed fields, and whether the agent should stop before submitting a change.

Chrome extension or built-in browser?

Use Chrome when…Use the built-in browser when…
The task needs your existing Chrome login, cookies, open tabs, or extensionsYou are working with public pages or want separate browser state
You need to continue from a page already open in your profileYou are testing localhost or a development preview
A site requires a session already authenticated in ChromeYou do not want the task to access your normal Chrome profile

ChatGPT may also use a dedicated app connection when one is enabled and better suited to the task. The current Chrome extension documentation explains this tool selection.

Use browser access safely

  • Check the active account before allowing actions on email, CRM, finance, or admin pages.
  • Treat instructions found on websites as untrusted; a page can contain malicious or misleading text.
  • Limit the task to the minimum sites and records required.
  • Review recipients, amounts, permissions, and final form values before submission.
  • Enter passwords and other credentials only in the browser, never in the chat.
  • Stop the task if the agent opens the wrong account or website.
  • Remove a site's approval or disconnect Chrome in settings when access is no longer needed.

Developer mode

Where available, Developer mode gives Codex controlled access to Chrome DevTools Protocol information for debugging console errors, network traffic, page state, and performance. It is disabled by default and should be enabled only for a development task that needs those details.

For other ways to use AI at work, compare these AI productivity tools. Browser control is most effective when the user defines the boundaries and remains responsible for important actions.

Discussion

Reader Comments 0

Sign in with email or Google to join the discussion.