CursorAI/CLI/Q/Can cursor-agent CLI accept image input? If so, how?

• Can CursorAI/CLI (cursor-agent) accept image input? If so, how?

  • My Notes

    • Tested it this morning in my beans log entry (see the garden for Person/codekiln/GitHub/beans); it worked just fine. I pointed the CLI to a path to the image on my machine prepended with the @ symbol.
    • One of the Frustrations I've encountered with it is the friction with using MacOS/Screenshots, which, these days, seem to save to the Documents folder with a long filename that contains spaces, including a non-standard narrow space utf character at the end of the filename. CursorAI struggled to find an image whose absolute path had been copied from yazi, and said that the image was not on disk until it wrote a python program on the fly to find it.
      • I really just wish I could paste in to the terminal the way I do with Claude/Code.
        • This morning, Cursor itself told me I could drag and drop images into the cursor-agent terminal, which I haven't tested.
          • TODO try dragging images into CursorAI/CLI

  • Answer

    • AI/Response
      • Yes. The Cursor Agent CLI accepts image input via file paths / @, headless prompt paths, and (on macOS/Linux) clipboard paste.
      • Interactive mode: Use @ to select files and folders (including images) to include in context. Same behavior as in the editor.
      • Headless/print mode: Put image file paths in the prompt. The agent reads files (including images) via tool calling and uses vision-capable models when needed.
      • Example (headless):

        • agent -p "Analyze this image and describe what you see: ./screenshot.png"

      • Example (multiple images):

        • agent -p "Compare these two images: ./before.png ./after.png"

      • Paths can be relative to the current working directory or absolute. Supported image formats include .png, .jpg, .gif, .webp, .svg. The agent can also read other binary/media files when you reference them in the prompt.
      • Paste image from clipboard? (as of 2026-08-26 Wed) Yes on macOS and Linux; not yet on Windows.
        • How: In an interactive agent / cursor-agent session, press Ctrl+V (not Cmd+V on macOS) to attach a clipboard image. Pasting a copied image file also resolves to the actual image.
        • When it landed: Cursor CLI changelog – April 2026 added “Image paste shortcuts” (Ctrl+V); May 7, 2026 added Linux clipboard paste on Wayland and X11; later entries fix drag-and-drop / macOS paste edge cases.
        • Caveats: Use Ctrl+V, not the terminal’s usual paste (Cmd+V / right-click). Some terminals or setups (Windows Terminal, WSL, SSH, tmux) may still fail or dump raw bytes — then fall back to a file path. Cursor staff (Aug 2026 forum): paste works on macOS/Linux; Windows remains a known gap with @path or drag-and-drop as workarounds.
        • Other image paths still work: @ / file paths in interactive mode; paths in -p prompts; --image in any session (changelog, March 2026).
      • Sources:
        • Answer/Official from Cursor CLI changelog (April 2026 image paste shortcuts; May 2026 Linux clipboard; later paste/DnD fixes; March 2026 --image)
        • Using Agent in CLI – Selecting context (@ for files; Input shortcuts page still omits Ctrl+V image paste)
        • Pasting images from clipboard doesn’t work (staff: Ctrl+V handler on macOS; user confirmed after update)
        • Pasting images not work on Window cursor cli (staff: works on macOS/Linux; Windows not yet)
        • Claude Code – Interactive mode (Ctrl+V / Alt+V image paste; similar gesture)

  • Related

    • CursorAI/CLI
    • AI/Coding/Tool/Q/How can I do fire and forget tasks in each CLI AI coding tool like Claude Code cursor-agent and codex?