René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"mcp_tool_to_langchain parameters conversion error","articleBody":"### Describe the Bug\n\n## Bug Description\n\n`mcp_tool_to_langchain` in `src/sap_cloud_sdk/agentgateway/converters.py` builds the\nPydantic args schema with every field typed as `str` (or `str | None`), regardless of\nthe actual type declared in the MCP tool's `input_schema`.\n\n## Root Cause\n\nLines 85–88 of `converters.py` iterate over property *keys* only and hard-code `str`:\n\n```python\nfields: dict[str, Any] = {\n    k: (str, ...) if k in required else (str | None, Field(default=None))\n    for k in properties          # ← values (the property schemas) are never read\n}\n```\n\nThe property schema values (e.g. `{\"type\": \"integer\"}`) are never consulted.\n\n## Impact\n\nWhen an LLM passes a non-string argument (e.g. `limit=10`, `active=true`), Pydantic\ncoerces it to a string (`\"10\"`, `\"true\"`) before forwarding it to `call_tool`. The MCP\nserver then rejects the call because it expects the native JSON type.\n\n## Affected types\n\nAll non-string JSON Schema types are affected: `integer`, `number`, `boolean`, `array`,\n`object`.\n\n## Reproduction\n\n```python\nfrom unittest.mock import AsyncMock\nfrom sap_cloud_sdk.agentgateway import MCPTool\nfrom sap_cloud_sdk.agentgateway.converters import mcp_tool_to_langchain\n\ntool = MCPTool(\n    name=\"search\",\n    server_name=\"s\",\n    description=\"d\",\n    input_schema={\n        \"type\": \"object\",\n        \"required\": [\"limit\"],\n        \"properties\": {\n            \"limit\": {\"type\": \"integer\"},\n            \"active\": {\"type\": \"boolean\"},\n        },\n    },\n    url=\"https://example.com/mcp\",\n)\n\nlc_tool = mcp_tool_to_langchain(tool, AsyncMock(), lambda: \"token\")\nfields = lc_tool.args_schema.model_fields\n\n# Both fields are `str` — should be `int` and `bool`\nprint(fields[\"limit\"].annotation)   # \u003cclass 'str'\u003e   ← bug\nprint(fields[\"active\"].annotation)  # str | None      ← bug\n```\n\n## Expected Fix\n\nMap JSON Schema `\"type\"` values to Python types before building the Pydantic model:\n\n| JSON Schema type | Python type |\n|------------------|-------------|\n| `\"string\"`       | `str`       |\n| `\"integer\"`      | `int`       |\n| `\"number\"`       | `float`     |\n| `\"boolean\"`      | `bool`      |\n| `\"array\"`        | `list`      |\n| `\"object\"`       | `dict`      |\n| *(missing/other)*| `Any`       |\n\n## Scope\n\nOnly `mcp_tool_to_langchain` in `agentgateway/converters.py` is affected. No other\nconverters exist in the codebase.\n\n\n### Steps to Reproduce\n\n1. Use mcp tool without string argument\n\n### Expected Behavior\n\n1. it should consider parameters other than string\n\n### Screenshots\n\n_No response_\n\n### Used Versions\n\n- Python version via `python --version`: ...\n- SAP Cloud SDK for Python version: ...\n- Framework version (if applicable, e.g., Flask, FastAPI): ...\n\n\u003cdetails\u003e\u003csummary\u003eInstalled packages via \u003ccode\u003epip list\u003c/code\u003e or \u003ccode\u003euv pip list\u003c/code\u003e\u003c/summary\u003e\n\n```\nPackage list here\n```\n\u003c/details\u003e\n\n\n### Code Examples\n\n```python\n# Your code here\n```\n\n\n### Stack Trace\n\n_No response_\n\n### Log File\n\n\u003cdetails\u003e\u003csummary\u003eLog file\u003c/summary\u003e\n...\n\u003c/details\u003e\n\n\n### Affected Development Phase\n\nGetting Started\n\n### Impact\n\nNo Impact\n\n### Timeline\n\n_No response_","author":{"url":"https://github.com/51acorsi","@type":"Person","name":"51acorsi"},"datePublished":"2026-06-26T20:57:55.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":2},"url":"https://github.com/190/cloud-sdk-python/issues/190"}