fix: Add pydantic complex types handling for MCP tools

[markindanila]

Currently MCP tools handle just simple types, which leads to agent not being able to provide the correct input for the complex tool when asked a question.

With this fix it's able to handle nested structures and Enums - see the example output of agent asked to show inputSchema

UPD:
list of changes in base.mcp.utils.create_input_schema_from_json_schema function:

• add Enums definitions parsing from $defs field of Pydantic JSON schema
• add BaseModel definitions recursive parsing from $defs field of Pydantic JSON schema
• add inheritance of Pydantic metadata fields

With this PR all types of BaseModel are in general supported including Enum attributes with these exceptions:

• attributes of type dict[<type1>, <type2>] are not parsed deeper than just dict since dict keys typing cannot be inherited from JSON Pydantic schema
• circular attribute references cannot be resolved properly, e.g. in this case both classes will most likely fallback to string:

class A(BaseModel):
    a: B

class B(BaseModel):
    b: A

[Copilot]

Copilot reviewed 1 out of 1 changed files in this pull request and generated no comments.

Comments suppressed due to low confidence (1)

src/backend/base/langflow/base/mcp/util.py:105

• Consider adding error handling to check that 'ref_name' exists in enum_defs before attempting to return it, to avoid a potential KeyError if an undefined reference is encountered.

return enum_defs[ref_name]

[Copilot]

Copilot reviewed 1 out of 1 changed files in this pull request and generated no comments.

Comments suppressed due to low confidence (3)

src/backend/base/langflow/base/mcp/util.py:109

• Ensure that enum values are strings before applying re.sub (e.g., by converting v to str) to avoid potential type errors.

enum_members = {re.sub(r"\W|^(?=\d)", "_", v): v for v in enum_def["enum"]}

src/backend/base/langflow/base/mcp/util.py:167

• [nitpick] The use of tuple[*item_types] may lead to compatibility issues with Python's generics syntax; consider using a supported syntax such as Tuple[...], ensuring compatibility with your target Python version.

return tuple[*item_types]  # type: ignore # noqa: PGH003

src/backend/base/langflow/base/mcp/util.py:188

• The updated check for required fields relies on a field-level 'required' key, but JSON schema normally specifies required fields at the schema level; consider reintroducing the schema-level required check to improve accuracy.

if "default" in field_def or not field_metadata["required"]:

[ogabrielluiz]

Hey @markindanila

Lookin good!

I'd say we need as many tests as we can on this.

[markindanila]

@ogabrielluiz Hi!
I've added the test here, not sure tho how to make the CI run it, but it works locally