Update mcp_client.py

mcp_client.py

@@ -27,6 +27,7 @@ class UniversalMCPClient:
         self.current_provider = None
         self.current_model = None
         self.server_tools = {}  # Cache for server tools
+        self.generation_params: Dict[str, Any] = {}
         
         # Initialize HF Inference Client if token is available
         if AppConfig.HF_TOKEN:
@@ -63,6 +64,13 @@ class UniversalMCPClient:
         self.current_provider = provider_id
         self.current_model = model_id
         logger.info(f"🔧 Set provider: {provider_id}, model: {model_id}")
+
+    def set_generation_params(self, params: Dict[str, Any]):
+        """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}
+        self.generation_params = cleaned
+        logger.info(f"🔧 Updated generation params: {list(self.generation_params.keys())}")
     
     def get_model_endpoint(self) -> str:
         """Get the current model endpoint for API calls"""
@@ -241,11 +249,16 @@ class UniversalMCPClient:
             "stream": kwargs.get("stream", False)
         }
         
-        # Add any remaining kwargs
+        # Merge stored generation params (do not override explicitly provided kwargs later)
+        for k, v in self.generation_params.items():
+            if k not in ("model", "messages") and k not in params:
+                params[k] = v
+
+        # Add any remaining kwargs (highest precedence)
         params.update(kwargs)
         
         # Add reasoning effort if specified (GPT OSS feature)
-        reasoning_effort = kwargs.pop("reasoning_effort", AppConfig.DEFAULT_REASONING_EFFORT)
+        reasoning_effort = kwargs.pop("reasoning_effort", self.generation_params.get("reasoning_effort", AppConfig.DEFAULT_REASONING_EFFORT))
         if reasoning_effort:
             # For GPT OSS models, we can set reasoning in system prompt
             system_message = None