fix: mcp_tool_to_langchain silently drops odata parameters whose name…

[NicoleMGomes]

Disclaimer: Do not include SAP-internal or customer-specific information in this PR (e.g. internal system URLs, customer names, tenant IDs, or confidential configurations). This is a public repository.

Description

mcp_tool_to_langchain in converters.py was calling pydantic.create_model with OData property names directly. Pydantic v2 rejects field names starting with _ (raised as NameError in newer versions, silently dropped in older ones). Because OData CSDL §15.2 explicitly allows _ as a valid first character for identifiers (e.g. _VariantConfiguration, _Product), any agent using mcp_tool_to_langchain against an OData MCP server with _-prefixed parameters received a broken tool schema — those parameters were invisible to the LLM and tool calls would fail or produce wrong results.

Important: passing the JSON Schema dict directly as args_schema does not work — LangChain internally converts any dict back to a Pydantic model, so the problem persists. The fix instead builds an internal name_map that strips leading underscores before passing field names to create_model, then restores the original OData names inside run() before forwarding kwargs to call_tool. This is transparent to callers.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update
• Code refactoring
• Dependency update

How to Test

1. Create an MCPTool whose input_schema contains a _-prefixed property:

   from sap_cloud_sdk.agentgateway import MCPTool
   from sap_cloud_sdk.agentgateway.converters import mcp_tool_to_langchain
   from unittest.mock import AsyncMock
   
   tool = MCPTool(
       name="get_variant_config",
       server_name="s4hana",
       description="Get variant configuration",
       input_schema={
           "type": "object",
           "required": ["_VariantConfiguration"],
           "properties": {
               "_VariantConfiguration": {"type": "string"},
               "_Product": {"type": "string"},
           },
       },
       url="https://example.com/mcp",
   )
   lc_tool = mcp_tool_to_langchain(tool, AsyncMock(), lambda: "token")

2. Verify "VariantConfiguration" (stripped) appears as a required field in lc_tool.args_schema.model_fields.
3. Invoke the tool and verify call_tool receives _VariantConfiguration (restored):

   import asyncio
   call_tool = AsyncMock(return_value="ok")
   lc_tool = mcp_tool_to_langchain(tool, call_tool, lambda: "token")
   asyncio.run(lc_tool.arun({"VariantConfiguration": "VC001"}))
   assert "_VariantConfiguration" in call_tool.call_args.kwargs

4. Run the unit tests: uv run pytest tests/agentgateway/unit/test_converters.py -v
5. All 29 tests should pass.

Checklist

• I have read the Contributing Guidelines
• I have verified that my changes solve the issue
• I have added/updated automated tests to cover my changes
• All tests pass locally
• I have verified that my code follows the Code Guidelines
• I have updated documentation (if applicable)
• I have added type hints for all public APIs
• My code does not contain sensitive information (credentials, tokens, etc.)
• I have followed Conventional Commits for commit messages

Additional Notes

Teams affected: any agent using mcp_tool_to_langchain against an OData MCP server (e.g. S/4HANA, EWM). The EWM warehouse agent shipped a local _sanitize_mcp_tool_schema workaround for this; that workaround is now redundant.