CallToolResult structuredContent is not alias-normalized to match outputSchema

Comece em mcp/server/mcpserver/utilities/func_metadata.py, em FuncMetadata.convert_result, e compare o branch de CallToolResult com o caminho de retorno simples descrito na issue. Execute primeiro o exemplo de alias fornecido; estará concluído quando CallToolResult.structured_content usar os aliases de outputSchema após a conversão, preservando o restante do resultado.

Initial Checks

• I confirm that I'm using the newest release of my line (the latest 2.x, or the latest 1.x if I'm still on v1)
• I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue

Release line

2.x (current stable)

Description

FuncMetadata.convert_result treats a plain model return and a CallToolResult return differently for Pydantic field aliases.

• Plain returns: validate_python(..., by_alias=True, by_name=True) then model_dump(..., by_alias=True) / dump_python(..., by_alias=True), so structuredContent keys match outputSchema.
• CallToolResult path: validate_python(result.structured_content) only, then returns the result unchanged.

Because validation accepts Python field names when populate_by_name / by_name is in play, a tool can pass validation while emitting wire keys that do not match the published outputSchema (which uses aliases). Clients that validate structuredContent against outputSchema then reject a server-produced payload.

Related but not the same as #1073 / #1099 (plain-return alias dump) or #3100 / #3118 (schema validation vs serialization mode). This is specifically the CallToolResult short-circuit skipping alias normalization.

Verified on current main tip 08a3bc8eaf5bb69a6cb05ac708a86f5325977c20.

Expected: after validating CallToolResult.structured_content, normalize it the same way as the plain-return path (by_alias dump) before returning.

Example Code

from __future__ import annotations
import json
from typing import Annotated

from pydantic import BaseModel, ConfigDict, Field
from mcp_types import CallToolResult, TextContent
from mcp.server.mcpserver.utilities.func_metadata import func_metadata


class AliasOut(BaseModel):
    model_config = ConfigDict(populate_by_name=True)
    field_first: str = Field(alias="first")
    field_second: str = Field(alias="second")


def plain() -> AliasOut:
    return AliasOut(field_first="a", field_second="b")


def via_call_tool_result() -> Annotated[CallToolResult, AliasOut]:
    # Python field names — accepted by validate when populate_by_name is on
    return CallToolResult(
        content=[TextContent(text="ok")],
        structured_content={"field_first": "a", "field_second": "b"},
    )


plain_meta = func_metadata(plain)
ctr_meta = func_metadata(via_call_tool_result)

print("outputSchema keys:", sorted(plain_meta.output_schema["properties"]))
print("plain:", json.dumps(plain_meta.convert_result(plain()).structured_content, sort_keys=True))
print(
    "CallToolResult:",
    json.dumps(ctr_meta.convert_result(via_call_tool_result()).structured_content, sort_keys=True),
)

Observed on main @ 08a3bc8:

outputSchema keys: ['first', 'second']
plain: {"first": "a", "second": "b"}
CallToolResult: {"field_first": "a", "field_second": "b"}

Python & MCP Python SDK

• Python: 3.12.10
• MCP Python SDK: current main @ 08a3bc8eaf5bb69a6cb05ac708a86f5325977c20 (2.1.2.dev16+08a3bc8)
• Pydantic: 2.12.5 (venv) / also reproduced with system 2.13.5 for imports

AI disclosure

This issue was drafted with AI assistance (Cursor / Grok). I verified the reproduction on current main and reviewed the convert_result paths in func_metadata.py before filing. I would like to open a fix PR for this issue.