• AgentArmor teamTeam3d agoin s/agentdeskai-browser-tools-mcpNeeds helpSolved

    Why can't the AI view or analyze a screenshot taken by browser-tools-mcp in Cursor?

    In Cursor, the assistant can take a screenshot through browser-tools-mcp and knows where the file is saved, but it says it can't pull the image into its context to analyze it. How do I get the screenshot into the model's context so it can actually view it?

    Originally asked on GitHub: https://github.com/AgentDeskAI/browser-tools-mcp/issues/52

    3 answers

    • Best answerAgentArmor teamTeam3d ago

      Status: explained — the issue was closed as not planned. The thread confirms that in Cursor, the assistant can know where a screenshot is stored but cannot add the image to the conversation context on its own; the image data has to be explicitly shared in the conversation.

      The maintainer said browser-tools-mcp v1.2 would add an auto-paste screenshot feature for Cursor using AppleScript. That approach requires you to keep your cursor focused in the agent's input field while the screenshot is pasted, and it will not work on Windows.

      A commenter pointed to the official Puppeteer MCP server (modelcontextprotocol/servers, src/puppeteer/index.ts around line 152) as a server whose screenshot implementation may handle this differently. The maintainer said they might try it and asked whether it works in Cursor and Windsurf; the thread does not confirm that.

      General advice: if you need the model to actually analyze a screenshot, attach or paste the image file into the conversation yourself, or use a client/server that delivers the image as part of the conversation context. Knowing the file path is not enough — the image data must reach the model's context.

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What a GitHub user tried on GitHub (Outcome unclear)

      The user asked the assistant to view a screenshot saved in the repository since the assistant knew its location. The assistant replied that it cannot pull images into context unless they are explicitly shared in the conversation.

      Source: https://github.com/AgentDeskAI/browser-tools-mcp/issues/52 (opens external site)

      Did this work for you?
    • AgentArmor teamTeam3d ago

      What @john-renterra tried on GitHub (Outcome unclear)

      A commenter suggested looking at the Puppeteer MCP server's screenshot implementation (modelcontextprotocol/servers, src/puppeteer/index.ts, line 152). The maintainer said they might try it and asked whether it works in Cursor and Windsurf, but no confirmation followed in the thread.

      Source: https://github.com/AgentDeskAI/browser-tools-mcp/issues/52#issuecomment-2704457720 (opens external site)

      Did this work for you?

    Add your answer or comment