Problem
When DOMShell drives the browser, there is no separation between the agent's tabs and the user's tabs — the agent attaches to a tab inside the user's normal Chrome window. You cannot tell which tabs the agent controls, and nothing stops it from navigating or acting on a tab you are actively using. Sharing one Chrome window between a human and an agent is error-prone.
Proposal
When an MCP session starts, DOMShell creates a labeled Chrome tab group and places its working tab(s) inside it. Every shell command is confined to that group — a command targeting a tab outside the group is rejected. The agent drives its tab in the background (CDP works on inactive tabs without stealing focus), so you can browse, click, and navigate freely in any tab outside the group, in the same window.
This mirrors the tab-group pattern used by Claude for Chrome.
Scope (v1)
- In-window Chrome tab group — not a separate window or Chrome profile.
- Shares your existing Chrome session and logins — isolation is a workspace lane, not a separate identity.
- A single active session, enforced.
- Non-destructive teardown — the agent never closes your tabs.
Acceptance
- Starting a session produces a visible, labeled tab group containing the agent's tab(s).
- Commands targeting a tab outside the group are rejected.
- Navigating or clicking in tabs outside the group is unaffected by the agent.
- The agent never steals your foreground — tabs it opens are created inactive.
- Ending a session leaves your tabs intact.
From roadmap: Agent Ergonomics
Problem
When DOMShell drives the browser, there is no separation between the agent's tabs and the user's tabs — the agent attaches to a tab inside the user's normal Chrome window. You cannot tell which tabs the agent controls, and nothing stops it from navigating or acting on a tab you are actively using. Sharing one Chrome window between a human and an agent is error-prone.
Proposal
When an MCP session starts, DOMShell creates a labeled Chrome tab group and places its working tab(s) inside it. Every shell command is confined to that group — a command targeting a tab outside the group is rejected. The agent drives its tab in the background (CDP works on inactive tabs without stealing focus), so you can browse, click, and navigate freely in any tab outside the group, in the same window.
This mirrors the tab-group pattern used by Claude for Chrome.
Scope (v1)
Acceptance
From roadmap: Agent Ergonomics