AWS Nova Canvas MCP Server

Instructions

Inpaint a specific part of an image using a text mask prompt.

Args:
    image_path: File path of the original image
    prompt: Text prompt for the area to be inpainted
    mask_prompt: Text prompt for specifying the area to be masked (e.g., "window", "car")
    negative_prompt: Text prompt for excluding attributes from generation
    height: Output image height (pixels)
    width: Output image width (pixels)
    cfg_scale: Image matching degree for the prompt (1-20)
    open_browser: Whether to open the image in the browser after generation
    
Returns:
    Dict: Dictionary containing the file path of the inpainted image

Implementation Reference

• aws_nova_canvas_mcp/tools/inpainting.py:12-77 (handler)

  The main handler function for the 'inpainting' tool. It reads the input image, encodes it to base64, constructs a JSON payload with inpainting parameters for the Bedrock model, generates the image, saves it using image_storage utilities, and returns the saved image path.

  async def inpainting(
          image_path: str,
          prompt: str,
          mask_prompt: str,
          negative_prompt: str = "",
          height: int = 512,
          width: int = 512,
          cfg_scale: float = 8.0,
          open_browser: bool = True,
          output_path: str = None,
  ) -> Dict[str, Any]:
      """
      Inpaint a specific part of an image using a text mask prompt.
      
      Args:
          image_path: File path of the original image
          prompt: Text prompt for the area to be inpainted
          mask_prompt: Text prompt for specifying the area to be masked (e.g., "window", "car")
          negative_prompt: Text prompt for excluding attributes from generation
          height: Output image height (pixels)
          width: Output image width (pixels)
          cfg_scale: Image matching degree for the prompt (1-20)
          open_browser: Whether to open the image in the browser after generation
          output_path: Absolute path to save the image
          
      Returns:
          Dict: Dictionary containing the file path of the inpainted image
      """
      try:
          # Read image file and encode to base64
          with open(image_path, "rb") as image_file:
              input_image = base64.b64encode(image_file.read()).decode('utf8')
  
          body = json.dumps({
              "taskType": "INPAINTING",
              "inPaintingParams": {
                  "text": prompt,
                  "negativeText": negative_prompt,
                  "image": input_image,
                  "maskPrompt": mask_prompt
              },
              "imageGenerationConfig": {
                  "numberOfImages": 1,
                  "height": height,
                  "width": width,
                  "cfgScale": cfg_scale
              }
          })
  
          # Generate image
          image_bytes = generate_image(body)
  
          # Save image
          image_info = save_image(image_bytes, open_browser=open_browser, output_path=output_path)
  
          # Generate result
          result = {
              "image_path": image_info["image_path"],
              "message": f"Inpainting completed successfully. Saved location: {image_info['image_path']}"
          }
  
          return result
  
      except Exception as e:
          raise McpError(f"Error occurred while inpainting: {str(e)}")

• aws_nova_canvas_mcp/server.py:32-32 (registration)

  Location where the inpainting tool is imported and registered with the MCP server (registration is currently commented out).

  # mcp.add_tool(inpainting)