fix: Return images as MCP image resources instead of text

Changed MCP tools/call response format to properly return images:
- Detects when result contains image data (image or dataUrl field)
- Extracts base64 data from data URI
- Returns as MCP image resource with type='image' and mimeType='image/png'
- Also includes text description of the operation
- Non-image results still returned as formatted JSON text

This fixes HuggingChat not displaying thumbnail images.
Now HuggingChat will show the actual PNG image instead of just JSON.

mcp_server_comprehensive.py

@@ -665,16 +665,65 @@ async def mcp_post(request: Request):
             # Execute the tool
             result = await TOOLS[tool_name](arguments)
 
+            # Check if result contains an image
+            content_items = []
+
+            if result.get("success") and result.get("image"):
+                # Add image as image content
+                image_data = result["image"]
+
+                # Extract base64 data (remove data:image/png;base64, prefix if present)
+                if image_data.startswith("data:"):
+                    image_data = image_data.split(",", 1)[1]
+
+                content_items.append({
+                    "type": "image",
+                    "data": image_data,
+                    "mimeType": "image/png"
+                })
+
+                # Add text description
+                description = f"Successfully created thumbnail ({result.get('width')}x{result.get('height')}px)"
+                if result.get("steps"):
+                    description += f"\n\nSteps completed:\n"
+                    for step in result["steps"]:
+                        step_name = step.get("step", "").replace("_", " ").title()
+                        description += f"- {step_name}\n"
+
+                content_items.append({
+                    "type": "text",
+                    "text": description
+                })
+            elif result.get("success") and result.get("dataUrl"):
+                # Handle canvas_export result format
+                image_data = result["dataUrl"]
+
+                # Extract base64 data
+                if image_data.startswith("data:"):
+                    image_data = image_data.split(",", 1)[1]
+
+                content_items.append({
+                    "type": "image",
+                    "data": image_data,
+                    "mimeType": "image/png"
+                })
+
+                content_items.append({
+                    "type": "text",
+                    "text": f"Canvas exported successfully ({result.get('width')}x{result.get('height')}px)"
+                })
+            else:
+                # Regular text response for non-image results
+                content_items.append({
+                    "type": "text",
+                    "text": json.dumps(result, indent=2)
+                })
+
             return {
                 "jsonrpc": "2.0",
                 "id": request_id,
                 "result": {
-                    "content": [
-                        {
-                            "type": "text",
-                            "text": json.dumps(result)
-                        }
-                    ]
+                    "content": content_items
                 }
             }