diff --git a/letta/llm_api/openai_client.py b/letta/llm_api/openai_client.py
index 991e8d84..48201362 100644
--- a/letta/llm_api/openai_client.py
+++ b/letta/llm_api/openai_client.py
@@ -620,12 +620,20 @@ class OpenAIClient(LLMClientBase):
 
         client = OpenAI(**self._prepare_client_kwargs(llm_config))
         # Route based on payload shape: Responses uses 'input', Chat Completions uses 'messages'
-        if "input" in request_data and "messages" not in request_data:
-            resp = client.responses.create(**request_data)
-            return resp.model_dump()
-        else:
-            response: ChatCompletion = client.chat.completions.create(**request_data)
-            return response.model_dump()
+        try:
+            if "input" in request_data and "messages" not in request_data:
+                resp = client.responses.create(**request_data)
+                return resp.model_dump()
+            else:
+                response: ChatCompletion = client.chat.completions.create(**request_data)
+                return response.model_dump()
+        except json.JSONDecodeError as e:
+            logger.error(f"[OpenAI] Failed to parse API response as JSON: {e}")
+            raise LLMServerError(
+                message=f"OpenAI API returned invalid JSON response (likely an HTML error page): {str(e)}",
+                code=ErrorCode.INTERNAL_SERVER_ERROR,
+                details={"json_error": str(e), "error_position": f"line {e.lineno} column {e.colno}"},
+            )
 
     @trace_method
     async def request_async(self, request_data: dict, llm_config: LLMConfig) -> dict:
@@ -638,12 +646,20 @@ class OpenAIClient(LLMClientBase):
         kwargs = await self._prepare_client_kwargs_async(llm_config)
         client = AsyncOpenAI(**kwargs)
         # Route based on payload shape: Responses uses 'input', Chat Completions uses 'messages'
-        if "input" in request_data and "messages" not in request_data:
-            resp = await client.responses.create(**request_data)
-            return resp.model_dump()
-        else:
-            response: ChatCompletion = await client.chat.completions.create(**request_data)
-            return response.model_dump()
+        try:
+            if "input" in request_data and "messages" not in request_data:
+                resp = await client.responses.create(**request_data)
+                return resp.model_dump()
+            else:
+                response: ChatCompletion = await client.chat.completions.create(**request_data)
+                return response.model_dump()
+        except json.JSONDecodeError as e:
+            logger.error(f"[OpenAI] Failed to parse API response as JSON: {e}")
+            raise LLMServerError(
+                message=f"OpenAI API returned invalid JSON response (likely an HTML error page): {str(e)}",
+                code=ErrorCode.INTERNAL_SERVER_ERROR,
+                details={"json_error": str(e), "error_position": f"line {e.lineno} column {e.colno}"},
+            )
 
     def is_reasoning_model(self, llm_config: LLMConfig) -> bool:
         return is_openai_reasoning_model(llm_config.model)
diff --git a/letta/schemas/providers/base.py b/letta/schemas/providers/base.py
index 73e4a239..bad77164 100644
--- a/letta/schemas/providers/base.py
+++ b/letta/schemas/providers/base.py
@@ -263,6 +263,11 @@ class ProviderCreate(ProviderBase):
     base_url: str | None = Field(None, description="Base URL used for requests to the provider.")
     api_version: str | None = Field(None, description="API version used for requests to the provider.")
 
+    @field_validator("api_key", "access_key", mode="before")
+    @classmethod
+    def strip_whitespace(cls, v: str | None) -> str | None:
+        return v.strip() if isinstance(v, str) else v
+
 
 class ProviderUpdate(ProviderBase):
     api_key: str = Field(..., description="API key or secret key used for requests to the provider.")
@@ -271,6 +276,11 @@ class ProviderUpdate(ProviderBase):
     base_url: str | None = Field(None, description="Base URL used for requests to the provider.")
     api_version: str | None = Field(None, description="API version used for requests to the provider.")
 
+    @field_validator("api_key", "access_key", mode="before")
+    @classmethod
+    def strip_whitespace(cls, v: str | None) -> str | None:
+        return v.strip() if isinstance(v, str) else v
+
 
 class ProviderCheck(BaseModel):
     provider_type: ProviderType = Field(..., description="The type of the provider.")
@@ -279,3 +289,8 @@ class ProviderCheck(BaseModel):
     region: str | None = Field(None, description="Region used for requests to the provider.")
     base_url: str | None = Field(None, description="Base URL used for requests to the provider.")
     api_version: str | None = Field(None, description="API version used for requests to the provider.")
+
+    @field_validator("api_key", "access_key", mode="before")
+    @classmethod
+    def strip_whitespace(cls, v: str | None) -> str | None:
+        return v.strip() if isinstance(v, str) else v
diff --git a/letta/services/mcp/base_client.py b/letta/services/mcp/base_client.py
index d297e83f..6b4a0536 100644
--- a/letta/services/mcp/base_client.py
+++ b/letta/services/mcp/base_client.py
@@ -87,6 +87,9 @@ class AsyncBaseMCPClient:
             # Log at debug level to avoid triggering production alerts for expected failures
             if e.__class__.__name__ in ("McpError", "ToolError"):
                 logger.debug(f"MCP tool '{tool_name}' execution failed: {str(e)}")
+                # Return error message with failure status instead of raising to avoid Datadog alerts
+                return str(e), False
+            # Re-raise unexpected errors
             raise
 
         parsed_content = []