Claude in Chrome Now Carries Your Cowork Session Across Desktop, Web, and Mobile

On August 12, 2026, Anthropic turned the Claude in Chrome side panel into a full Claude Cowork session. That sounds like a small wording change, but it removes a real piece of friction anyone running Claude-based workflows daily has hit: until now, a Claude in Chrome session was isolated to the tab it started in. Close the browser, switch to your phone, and the work was gone. Claude in Chrome Cowork sessions now live with your account rather than a single device, so a task started in the browser can be picked up later on desktop, web, or mobile with its full context intact.

What actually changed

Previously, side panel conversations never touched Claude's history. They existed only inside that browser tab, on that machine, for that session. Anthropic's own announcement describes the new behavior plainly: conversations, skills, and connectors are now saved to your account and accessible from any Claude surface. Skills and connectors you already have configured elsewhere carry over into the browser automatically, with no separate setup step for the extension.

The practical shift is that the browser stops being a dead end for a Claude conversation. Anthropic's example is a budget spreadsheet task: Claude collects invoice data from several vendor portals inside the browser, using logins you already have open, then the session moves to the desktop app to handle the actual file uploads and comparative analysis. Before this update, that handoff meant manually re-explaining the task on the second device. Now the context transfers with it.

What Claude in Chrome can do on a page

The underlying browser capability itself is not new, but it is worth restating because it is what makes the cross-device piece useful rather than cosmetic. Claude in Chrome can see the page you are viewing, click links, navigate between pages, type into fields, and submit forms, all using your existing browser logins rather than a separate API connection. That matters for the class of tool most agent frameworks still struggle with: internal dashboards, legacy systems, and vendor portals that were never built with an API in the first place. If a human can click through it in a browser, Claude in Chrome can now act on it directly, and that session's progress is no longer stranded on whichever machine happened to be open when the task started.

The security layer

Letting an agent click, type, and submit inside your own logged-in browser session raises an obvious question, and Anthropic addressed it directly rather than leaving it implicit. The protection works in two tiers. With auto-approve on, Claude works through a task without stopping for permission at every step, but before anything consequential, meaning a form submission, a message, or a file download, a separate check reviews the action against what you originally asked for and blocks anything that does not match. A second, stricter tier sits above that: Claude still stops and asks before certain irreversible or costly actions specifically, like making a purchase or sharing personal data, regardless of auto-approve. Anthropic is also explicit that neither layer eliminates the risk: the company states that prompt injection, where a malicious page tries to redirect the agent's behavior through hidden or misleading content, remains an ongoing risk despite both checks. That is a more honest framing than most browser-agent launches offer, and it is worth taking at face value if you plan to point Claude in Chrome at anything with real financial or account access.

Rollout and what is still missing

The cross-device Cowork update is live now on Max and Team plans, with Pro rolling out over the coming weeks. Enterprise plans have it off by default, and admins can enable it restricted to a list of approved domains, which is the sensible default for any organization letting an agent act inside internal tools. Two real limitations remain: Claude in Chrome does not yet run on other Chromium-based browsers or on mobile browsers themselves, and local file work still requires the desktop app. The cross-device continuity covers the conversation and its context, not every capability Claude has.

What this means if you are building with Claude

If your daily workflow already spans a laptop, a phone, and a browser full of tools without clean APIs, this closes a specific gap rather than adding a new capability from scratch:

• A task that starts as browser research can finish as a desktop file operation, with the same session carrying the context, instead of you copying findings between two separate conversations.
• Skills and MCP connectors you have already built or installed elsewhere just work in the browser, without a second configuration pass specific to the extension.
• Anything reachable only through a login-gated web portal is now in scope for automation, not just services with a documented API, which changes the calculus for internal tools your team has never bothered to wrap in an API layer.
• The action-verification check is a real constraint, not a formality, worth testing directly against your own workflow before trusting it with anything that spends money or sends messages on your behalf, given Anthropic's own acknowledgment that prompt injection is not fully solved.

For anyone treating Claude as the coordination layer across several tools rather than a single chat window, the device a task starts on stops being a decision that matters. That is a small change in wording, but it removes a real piece of daily friction.

Sources: Claude by Anthropic, "Claude Cowork comes to the Chrome side panel", 9to5Mac, The New Stack