fix: pass MCP tool images to the model, not only to the UI (#30358)

When an MCP tool returns an image (e.g. a Home Assistant camera snapshot), the
snapshot shows up in the tool call section but the model never sees it: it
answers that there is no image. Only images arriving as inline data URIs were
attached to the model request; MCP images are uploaded to Files first and their
file URL was treated as display-only.

Now an image file item with a file URL is attached to the model request as an
input_image part in addition to staying in the tool call's displayed files. The
existing URL-to-base64 step already resolves file URLs, so the model receives
the image bytes; verified on a running instance with a mock MCP server and a
mock upstream (the second upstream request carries the byte-identical JPEG).
Inline data-URI images keep their existing model-only handling.

Fixes #30327

backend/open_webui/utils/middleware.py

@@ -3487,6 +3487,8 @@ async def drain_approved_tool_calls(request, form_data, user, model, metadata) -
                 output_parts.append({'type': 'input_image', 'image_url': image_url})
             else:
                 display_files.append(file_item)
+                if file_item.get('type') == 'image' and file_item.get('url'):
+                    output_parts.append({'type': 'input_image', 'image_url': file_item['url']})
 
         output.append(
             {
@@ -5858,7 +5860,7 @@ async def streaming_chat_response_handler(response, ctx):
                     frontend_output = []
                     for item in full_output() if snapshot else full_output()[output_start:]:
                         if item.get('type') == 'function_call_output':
-                            # input_image parts are base64 data URIs only for the LLM, via convert_output_to_messages
+                            # input_image parts are for the LLM only, via convert_output_to_messages
                             item = {
                                 **item,
                                 'output': [
@@ -6127,8 +6129,7 @@ async def streaming_chat_response_handler(response, ctx):
                         )
                         result_status_by_call_id[result.get('tool_call_id', '')] = local_output_status
 
-                        # Separate image data URIs (for LLM via input_image) from
-                        # other files (for frontend display via files attribute).
+                        # Data-URI images: LLM only. File-URL images: LLM and frontend. Other files: frontend.
                         display_files = []
                         for file_item in result.get('files', []):
                             if file_item.get('type') == 'image' and file_item.get('url', '').startswith('data:'):
@@ -6136,8 +6137,9 @@ async def streaming_chat_response_handler(response, ctx):
                                 image_url = await store_tool_result_image(request, file_item['url'], metadata, user)
                                 output_parts.append({'type': 'input_image', 'image_url': image_url})
                             else:
-                                # Frontend display (MCP images, audio, etc.)
                                 display_files.append(file_item)
+                                if file_item.get('type') == 'image' and file_item.get('url'):
+                                    output_parts.append({'type': 'input_image', 'image_url': file_item['url']})
 
                         output.append(
                             {