Playwright `screenshot` tool fills context window

[Burry]

The Playwright screenshot tool seems to fill the model's context window with base64-encoded image data. Could this data be omitted from the tool response, with the model directed to just read the screenshot image file?

⏺ next-devtools - browser_eval (MCP)(action: "screenshot")
  ⎿  ⚠ Large MCP response (~25.1k tokens), this can fill up context quickly         
  ⎿ {"success":true,"action":"screenshot","result":{"content":[{"type":"text","text"
    :"### Result\nTook the viewport screenshot and saved it as
    /var/folders/zn/3jf0y5g95pj4fhft836hf5p80000gp/T/playwright-mcp-output/176531185
    4814/page-2025-12-09T20-25-00-990Z.png\n\n### Ran Playwright code\n// Screenshot
     viewport and save it as /var/folders/zn/3jf0y5g95pj4fhft836hf5p80000gp/T/playwr
    ight-mcp-output/1765311854814/page-2025-12-09T20-25-00-990Z.png\nawait
    page.screenshot({\nscale: 'css',\n  type:
    'png'\n});\n"},{"type":"image","data":"iVBORw0KGgoAAAANSUhEUgAABCgAAARtCAYAAACX/
    [...]