mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-11 03:38:38 +00:00
6cc81584b9
JSON Schema allows "type" to be an array (e.g. ["string", "null"]) for a nullable/optional field, the shape Pydantic's model_json_schema() emits for Optional[str] / str | None. A real Azure DevOps MCP tool schema (testplan_test_suite_write) declares its optional parameters exactly this way. Both ToolTestPanel (the MCP Tool Testing Playground) and MCPToolArgumentsForm (the chat playground's tool-call argument form) compare a property's type with strict equality against a single string, with no fallback for the array case. In ToolTestPanel this leaves the field with no input at all, since none of the type checks match and there is no default branch: the field's label renders with nothing to type into, and the same strict comparison in the submit-time value conversion would also skip type coercion for it. MCPToolArgumentsForm has a final else branch, so the same schema shape instead renders the wrong widget (a plain text input for what should be a number or boolean field). Add resolveSchemaType to both files, resolving to the first non-"null" type when given an array, and use it everywhere a property's type decides a widget or a value conversion. Widen InputSchemaProperty.type in the shared types module to string | string[] to match. |
||
|---|---|---|
| .. | ||
| litellm-dashboard | ||
| Dockerfile | ||
| nginx.conf | ||