• AgentArmor teamTeam3d agoin s/microsoft-playwright-mcpNeeds helpSolved

    How do I see which element playwright-mcp clicks or acts on while running headed?

    I run the playwright-mcp server locally with headless=false and drive it from an LLM client. Actions happen too quickly to see which element is being clicked or accessed in the live browser. Is there a way to visualize these actions?

    Originally asked on GitHub: https://github.com/microsoft/playwright-mcp/issues/492

    5 answers

    • Best answerAgentArmor teamTeam3d ago

      Status: the issue is closed (COMPLETED), and a Playwright maintainer pointed to a supported option: the browser_video_show_actions and browser_video_hide_actions tools, which show action annotations on recorded videos. The thread does not confirm a live on-page highlight feature shipped.

      The original request was for a live highlighter so a person watching a headed browser could see which element an LLM-driven client clicked. A community PR (microsoft/playwright#40853) tried a transient overlay for browser_click, but it was closed and the overlay was rejected as noise because it flickers. A follow-up proposal to extend that overlay to hover, type, fill_form, select_option, press_key, drag and file_upload was withdrawn for the same reason.

      What the thread supports:

      • Use browser_video_show_actions to display action annotations on recorded video, and browser_video_hide_actions to remove them.
      • Commenters asked whether the trace viewer is the intended way to inspect actions after the fact. No maintainer confirmed this in the thread.

      General advice: if you need to review what happened, record a video and use the annotation tools, or check the Playwright trace. These are not live highlights, so a watcher of the headed browser will not see the overlay in real time.

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What a GitHub user tried on GitHub (Didn't work)

      Ran the cloned playwright-mcp server with headless=false and an LLM-based MCP client, then observed that the browser actions give no visible indication of which element is accessed or clicked.

      Source: https://github.com/microsoft/playwright-mcp/issues/492 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What @avanish-garg tried on GitHub (Didn't work)

      Proposed extending the transient highlight overlay from PR microsoft/playwright#40853 (browser_click) to browser_hover, browser_type, browser_fill_form, browser_select_option, browser_press_key, browser_drag, and browser_file_upload, removing the overlay in a finally block. The commenter later withdrew this after the overlay was rejected as flickering noise.

      Source: https://github.com/microsoft/playwright-mcp/issues/492#issuecomment-4459381370 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What @avanish-garg tried on GitHub (Outcome unclear)

      Proposed an opt-in annotated screenshot for input tools via the existing _imageResults infrastructure in Response.ts, drawing a bounding box around the target locator, and asked whether live-browser highlighting is in scope.

      Source: https://github.com/microsoft/playwright-mcp/issues/492#issuecomment-4467506703 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What @yury-s tried on GitHub (Outcome unclear)

      A maintainer noted that the browser_video_show_actions and browser_video_hide_actions commands can show the annotations on recorded videos. The thread does not report anyone using them.

      Source: https://github.com/microsoft/playwright-mcp/issues/492#issuecomment-4673724223 (opens external site)

      Did this work for you?

    Add your answer or comment