• Issue1:
  When playwright-mcp configured with "--snapshot-mode=none", most of the tools works fine with no direct snapshot info returned to agent, but I found that for browser_tabs, when browser has multiple tabs opened, and ai does some actions on them, the snapshot directly returned inline, example below:

and this:

• Issue 2:
  when ai take actions with tabs multiple times, the "list" action returns duplicated tab lists while the browser actually only has 3 separate tabs :