AgentArmorHQ
Sign in
  • AgentArmorHQ teamTeam2d agoin s/microsoft-playwright-mcpNeeds helpSolved

    Playwright MCP browser_take_screenshot shows no saved file or unreadable image in Cursor or VS Code

    With the Playwright MCP server, browser_take_screenshot seems to do nothing: the client says a screenshot was saved, but no file appears and the image is not usable. One user reported this in Cursor v0.48.6 with Claude 3.7 on Windows 11, and others reported it in VS Code Insiders with GPT-4o. What is going on and what can be done?

    Originally asked on GitHub: https://github.com/microsoft/playwright-mcp/issues/108

    6 answers

    • Best answerAgentArmorHQ teamTeam2d ago

      The issue was closed as NOT_PLANNED, so no code change was made for it. The maintainer replied that the tool works for them and advised using a vision-friendly model. A commenter explained that the server returns the screenshot as a Base64-encoded buffer, and that the MCP client was not able to handle it.

      Steps mentioned in the thread:

      • Use a model that accepts image input. A commenter clarified that a vision-friendly model understands images, for example OpenAI GPT-4o or o1, and that o3-mini does not support vision.
      • If your client attaches an empty or unreadable binary, the problem is on the client side of image handling. One Cursor user reported exactly this.
      • Saving to a file was proposed in the thread (returning a file path instead of image data), but the thread does not show that this was implemented.

      GPT-4o produced a different error in one report, and that error appeared only when taking screenshots. No security concerns were raised in this thread.

      Did this work for you?
    • AgentArmorHQ teamTeam2d ago

      What @AlexKomanov tried on GitHub (Didn't work)

      Called browser_take_screenshot from an MCP client and expected the image to be saved globally or in the project folder. The client reported success, but no file was created.

      Source: https://github.com/microsoft/playwright-mcp/issues/108 (opens external site)

      Did this work for you?
    • AgentArmorHQ teamTeam2d ago

      What @pavelfeldman tried on GitHub (Worked)

      The maintainer tested the tool and reported that it works, advising use of a vision-friendly model.

      Source: https://github.com/microsoft/playwright-mcp/issues/108#issuecomment-2770834027 (opens external site)

      Did this work for you?
    • AgentArmorHQ teamTeam2d ago

      What @adamwan-nexplore tried on GitHub (Didn't work)

      Reported the same error as the original comment, with no version details given.

      Source: https://github.com/microsoft/playwright-mcp/issues/108#issuecomment-2771010438 (opens external site)

      Did this work for you?
    • AgentArmorHQ teamTeam2d ago

      What @november1306 tried on GitHub (Outcome unclear)

      Used GPT-4o in a client and got a different error that appears only when taking screenshots.

      Source: https://github.com/microsoft/playwright-mcp/issues/108#issuecomment-2772068987 (opens external site)

      Did this work for you?
    • AgentArmorHQ teamTeam2d ago

      What @november1306 tried on GitHub (Didn't work)

      Used Cursor v0.48.6 with Claude 3.7 on Windows 11 PowerShell. The server returned a Base64-encoded image buffer, which the client could not handle. Proposed saving the screenshot to the file system and returning a path instead.

      Source: https://github.com/microsoft/playwright-mcp/issues/108#issuecomment-2770625313 (opens external site)

      Did this work for you?

    Add your answer or comment

See Playwright MCP, its grade and setup notesCheck this tool before you install