Update mcp_client.py

mcp_client.py

@@ -69,6 +69,10 @@ class UniversalMCPClient:
         """Set generation parameters for chat completions (OpenAI-compatible)."""
         # Clean None values to avoid sending invalid fields
         cleaned = {k: v for k, v in params.items() if v is not None}
+        # Defensive: If no tools provided, drop tool-specific params to avoid 400s
+        if "tools" not in cleaned:
+            cleaned.pop("tool_choice", None)
+            cleaned.pop("tool_prompt", None)
         self.generation_params = cleaned
         logger.info(f"🔧 Updated generation params: {list(self.generation_params.keys())}")
     
@@ -257,6 +261,11 @@ class UniversalMCPClient:
         # Add any remaining kwargs (highest precedence)
         params.update(kwargs)
         
+        # Defensive: If tools are absent, ensure tool-specific params are not sent
+        if "tools" not in params:
+            params.pop("tool_choice", None)
+            params.pop("tool_prompt", None)
+
         # Add reasoning effort only for GPT-OSS models
         if AppConfig.is_gpt_oss_model(self.current_model):
             reasoning_effort = kwargs.pop("reasoning_effort", self.generation_params.get("reasoning_effort", AppConfig.DEFAULT_REASONING_EFFORT))
@@ -287,6 +296,62 @@ class UniversalMCPClient:
         except Exception as e:
             logger.error(f"HF Inference API call failed: {e}")
             raise
+
+    def generate_chat_completion_stream(self, messages: List[Dict[str, Any]], **kwargs):
+        """Stream chat completion tokens using HuggingFace Inference Providers (OpenAI-compatible)."""
+        if not self.hf_client:
+            raise ValueError("HuggingFace client not initialized. Please set HF_TOKEN.")
+
+        if not self.current_provider or not self.current_model:
+            raise ValueError("Provider and model must be set before making API calls")
+
+        model_endpoint = self.get_model_endpoint()
+
+        # Base params
+        params = {
+            "model": model_endpoint,
+            "messages": messages,
+            "max_tokens": kwargs.pop("max_tokens", 8192),
+            "temperature": kwargs.get("temperature", 0.3),
+            "stream": True,
+        }
+
+        # Merge stored generation params (exclude stream keys)
+        for k, v in self.generation_params.items():
+            if k not in ("model", "messages", "stream") and k not in params:
+                params[k] = v
+
+        # Add any remaining kwargs (highest precedence)
+        params.update(kwargs)
+
+        # Defensive: If tools are absent, ensure tool-specific params are not sent
+        if "tools" not in params:
+            params.pop("tool_choice", None)
+            params.pop("tool_prompt", None)
+
+        # Apply reasoning only for GPT-OSS models
+        if AppConfig.is_gpt_oss_model(self.current_model):
+            reasoning_effort = kwargs.pop("reasoning_effort", self.generation_params.get("reasoning_effort", AppConfig.DEFAULT_REASONING_EFFORT))
+            if reasoning_effort:
+                system_message = None
+                for msg in messages:
+                    if msg.get("role") == "system":
+                        system_message = msg
+                        break
+                if system_message:
+                    system_message["content"] += f"\n\nReasoning: {reasoning_effort}"
+                else:
+                    messages.insert(0, {"role": "system", "content": f"You are a helpful AI assistant. Reasoning: {reasoning_effort}"})
+        else:
+            params.pop("reasoning_effort", None)
+
+        try:
+            logger.info(f"🤖 Streaming from {model_endpoint} via {self.current_provider}")
+            stream = self.hf_client.chat.completions.create(**params)
+            return stream
+        except Exception as e:
+            logger.error(f"HF Inference API streaming failed: {e}")
+            raise
     
     def generate_chat_completion_with_mcp_tools(self, messages: List[Dict[str, Any]], **kwargs) -> Dict[str, Any]:
         """Generate chat completion with MCP tool support"""