feat: Enable MCP

app.py

@@ -19,6 +19,18 @@ generation_config_multi = model_multi.default_generation_config
 # MULTI-TURN INFERENCE FUNCTION
 # ---------------------------------
 def multi_turn_chat(user_input, audio_file, history, current_audio):
+    """
+    Handle multi-turn chat interactions with audio context.
+    
+    Args:
+        user_input: The user's text message/question about the audio
+        audio_file: New audio file path if uploaded, otherwise None
+        history: List of previous conversation turns as (user_msg, bot_response) tuples
+        current_audio: Path to the currently active audio file in the conversation
+        
+    Returns:
+        Tuple of (updated_chatbot_display, updated_history, updated_current_audio)
+    """
     try:
         if audio_file is not None:
             current_audio = audio_file  # Update state if a new file is uploaded
@@ -37,6 +49,15 @@ def multi_turn_chat(user_input, audio_file, history, current_audio):
         history.append((user_input, f"❌ Error: {str(e)}"))
         return history, history, current_audio
 def speech_prompt_infer(audio_prompt_file):
+    """
+    Process speech/audio input and generate a text response.
+    
+    Args:
+        audio_prompt_file: Path to the audio file containing the user's speech prompt
+        
+    Returns:
+        String containing the model's text response or error message
+    """
     try:
         sound = llava.Sound(audio_prompt_file)
         full_prompt = "<sound>"
@@ -197,4 +218,4 @@ To enable these capabilities, we propose several large-scale training datasets c
 # Launch App
 # -----------------------
 if __name__ == "__main__":
-    demo.launch(share=True)
+    demo.launch(share=True, mcp_server=True)