OpenAI releases Codex Chrome plugin

Browser access is where a lot of real work actually happens. Not in a demo sandbox — in Gmail, Salesforce, dashboards, staging sites, admin panels, and internal tools that all depend on your existing login state. That was the gap for Codex. OpenAI’s coding agent could already work in the cloud and inside its own app, but the in-app browser was cut off from the messy, useful parts of the web. This week, OpenAI closed a big chunk of that gap with a Chrome extension for macOS and Windows. ### What actually shipped? OpenAI added a Codex Chrome extension that lets Codex use the Chrome browser on your own computer. The point is simple: if a task depends on your signed-in browser session, Codex can now work there instead of getting stuck in a clean-room browser with no cookies, no auth, and no tab history. OpenAI published the extension docs and listed it in the Chrome Web Store on May 7. (developers.openai.com) ### Why does Chrome matter so much? Because the old browser inside Codex had hard limits. It could open public pages, but it did not support authentication flows, signed-in pages, your normal browser profile, extensions, or your existing tabs. That made it fine for public docs and basic testing, but weak for the kind of developer and ops work that lives behind logins. The Chrome extension changes that by letting Codex operate where your actual work already is. (developers.openai.com) ### What can Codex do there? OpenAI’s docs pitch very practical jobs — navigating structured sites, completing data-entry flows, checking dashboards, debugging browser behavior, and working with apps like LinkedIn, Salesforce, Gmail, or internal tools. The extension is meant for browser tasks that need your account state, not just page scraping. Basically, it turns Codex from “agent with a browser” into “agent in your browser.” (developers.openai.com) ### What’s the standout feature? Parallel work across tabs in the background. That sounds small, but it is the difference between an agent hijacking your screen and an agent acting more like a coworker doing side tasks. OpenAI says Codex can work across multiple Chrome tabs without taking over the browser, and related browser actions are grouped into Chrome tab groups tied to a specific Codex thread. (developers.openai.com) ### How much control does the user keep? More than the raw “computer use” framing suggests. OpenAI says users manage website approvals, allowlists, and blocklists, and stay in control of which websites Codex can use. The docs also frame Chrome as the preferred path when a plugin exists, instead of falling back to broader desktop control. So this is not just more power — it is also a more bounded way to grant that power. (developers.openai.com) ### Is this only about coding? Not really. Codex is still positioned as a coding agent, but the browser extension pushes it toward workflow automation more broadly. OpenAI’s app docs now bundle Chrome access alongside automations, skills, plugins, and background tasking. That combination matters because a lot of software work is half code and half browser glue — checking a dashboard, reproducing a bug, filing something in a tool, then going back to the repo. (developers.openai.com) ### What’s the catch? The catch is geography and trust. The Chrome Web Store listing is live, but at least one report says rollout starts outside the EU and UK, and the whole model depends on users being comfortable handing an agent access to authenticated web sessions one site at a time. That is useful — but it is also exactly where the risk gets real. (developers.openai.com) ### Bottom line? This is a product plumbing update, but it changes the shape of what Codex can actually do. The big unlock is not “Chrome support” by itself. It is that Codex can now work inside the logged-in browser surfaces where real developer and business workflows already live. (developers.openai.com) (chromewebstore.google.com)