## Vision support (deepseek-vl-support)

The main model is text-only (e.g. DeepSeek) and CANNOT see images. When you need to understand an image — screenshots, error screens, UI mockups, charts, diagrams — do NOT use `view_image` (text-only models cannot process images). Save the image to a file if needed, then call:

- `mcp__deepseek-vl__describe_image` with `{"path": "<image file path>", "question": "<optional focus>"}` → returns a detailed text description of the image (visible text, UI layout, colors, code, errors).
- `mcp__deepseek-vl__vision_status` → checks vision configuration and endpoint health (model visibility).

Notes:
- Pasted/dragged images are lost for text-only models: save the image as a file first, then describe it.
- If vision fails, report the failure and continue working with text.