Vision capabilities for crow-cli via MCP.

Capture a single frame from a webcam device.

Args:

• device_index: Webcam device index (default: 6)

Read an image from a file path for vision analysis.

Args:

• file_path: Absolute path to the image file (jpg, jpeg, png, bmp, etc.)

# Run the server
uv --project /home/thomas/src/crow-cli/crow-vision-mcp run python main.py

# Or with MCP client
uv run mcp-stdio -- server.py

# From a crow-cli agent
result = await mcp_client.call_tool("read_image_file", {"file_path": "/path/to/screenshot.png"})
# Returns image data that can be sent to vision-capable LLM

• JPEG/JPG
• PNG
• BMP (converted to JPEG for transmission)
• Any format OpenCV can read