fix: optional mcp tools

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sap-cloud-sdk"
-version = "0.27.0"
+version = "0.27.1"
 description = "SAP Cloud SDK for Python"
 readme = "README.md"
 license = "Apache-2.0"

src/sap_cloud_sdk/agentgateway/converters.py

@@ -74,7 +74,10 @@ async def run(**kwargs) -> str:
 
     # Build args schema from input_schema
     properties = mcp_tool.input_schema.get("properties", {})
-    fields: dict[str, Any] = {k: (str, ...) for k in properties}
+    required = set(mcp_tool.input_schema.get("required", []))
+    fields: dict[str, Any] = {
+        k: (str, ...) if k in required else (str | None, None) for k in properties
+    }
     args_schema = create_model(f"{mcp_tool.name}_args", **fields) if fields else None
 
     return StructuredTool.from_function(

tests/agentgateway/unit/test_converters.py

@@ -78,6 +78,34 @@ def test_handles_empty_input_schema(self):
         assert result.name == "simple_tool"
         assert result.args_schema is not None
 
+    def test_optional_fields_not_required_in_args_schema(self):
+        """Fields absent from 'required' must be optional in the generated Pydantic model."""
+        tool = MCPTool(
+            name="get_supplier_bid",
+            server_name="ariba",
+            description="Gets all supplier bids for the specified event",
+            input_schema={
+                "type": "object",
+                "required": ["eventid"],
+                "properties": {
+                    "eventid": {"description": "Unique identifier of the event"},
+                    "showdeclinedreason": {"description": "Show supplier decline reason"},
+                    "datafetchmode": {"description": "Level of detail for the response"},
+                },
+            },
+            url="https://example.com/mcp",
+        )
+
+        result = mcp_tool_to_langchain(tool, AsyncMock(return_value="result"), lambda: "token")
+
+        from pydantic import BaseModel
+
+        assert result.args_schema is not None and isinstance(result.args_schema, type) and issubclass(result.args_schema, BaseModel)
+        fields = result.args_schema.model_fields
+        assert fields["eventid"].is_required(), "eventid should be required"
+        assert not fields["showdeclinedreason"].is_required(), "showdeclinedreason should be optional"
+        assert not fields["datafetchmode"].is_required(), "datafetchmode should be optional"
+
     def test_handles_input_schema_without_properties(self):
         """Handle MCPTool with input schema but no properties."""
         tool = MCPTool(

tests/core/integration/telemetry/_agent.py

@@ -17,9 +17,9 @@ def build_langgraph_agent():
         from typing import Annotated
 
         from langchain_core.messages import BaseMessage
-        from langchain_litellm import ChatLiteLLM  # ty: ignore[unresolved-import]
-        from langgraph.graph import END, StateGraph  # ty: ignore[unresolved-import]
-        from langgraph.graph.message import add_messages  # ty: ignore[unresolved-import]
+        from langchain_litellm import ChatLiteLLM
+        from langgraph.graph import END, StateGraph
+        from langgraph.graph.message import add_messages
     except ImportError:
         pytest.skip("langchain-litellm or langgraph not installed")
 

tests/core/integration/telemetry/test_telemetry_bdd.py

@@ -127,7 +127,7 @@ def _llm_call():
     Uses ChatLiteLLM so Traceloop's LangChain instrumentor fires and emits a span
     with gen_ai.usage.* token counts.
     """
-    from langchain_litellm import ChatLiteLLM  # ty: ignore[unresolved-import]
+    from langchain_litellm import ChatLiteLLM
     from langchain_core.messages import HumanMessage as LCHumanMessage
     model_name = os.environ.get("AICORE_MODEL", "anthropic--claude-4.5-sonnet")
     llm = ChatLiteLLM(model=f"sap/{model_name}")