From a8aee9bb9f07322333c12fdb2b9e675c6307b4d3 Mon Sep 17 00:00:00 2001
From: LUCAS KIM <ichbinlucas211@gmail.com>
Date: Fri, 30 Jan 2026 18:57:55 -0800
Subject: [PATCH] fix: preserve non-ASCII text in MCP JSON serialization
 (#6995)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## Summary

This PR fixes MCP JSON serialization so that non-ASCII text (for example, Japanese) is preserved instead of being escaped as `\uXXXX`.

## Changes

- Added an internal helper `_to_json_str` in `autogen_ext.tools.mcp._base` that calls `json.dumps(..., ensure_ascii=False, indent=indent)`.
- Updated `McpToolAdapter.return_value_as_string` to use `_to_json_str` for serializing tool return values.
- Updated `StreamElicitor.elicit` in `_host/_elicitation.py` to use `_to_json_str` when rendering the requested schema for the user.
- Added `tests/tools/test_mcp_serialization.py` to verify that Japanese text is not escaped.

## Tests

- `uv run pytest tests/tools/test_mcp_serialization.py` ✓
- `uv run pytest` for `autogen-ext` fails on optional integrations (`chromadb`, `mem0`, `redisvl`, `torch`, `ollama`, `semantic_kernel`, `graphrag`, `json_schema_to_pydantic`) that are not installed in my local environment, but the new MCP tests pass.
---
 .../src/autogen_ext/tools/mcp/_base.py        |  6 +++-
 .../tools/mcp/_host/_elicitation.py           |  4 ++-
 .../tests/tools/test_mcp_serialization.py     | 34 +++++++++++++++++++
 3 files changed, 42 insertions(+), 2 deletions(-)
 create mode 100644 python/packages/autogen-ext/tests/tools/test_mcp_serialization.py

diff --git a/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_base.py b/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_base.py
index 314dbe15e7ba..cad75abb6361 100644
--- a/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_base.py
+++ b/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_base.py
@@ -194,4 +194,8 @@ def serialize_item(item: Any) -> dict[str, Any]:
             else:
                 return {}
 
-        return json.dumps([serialize_item(item) for item in value])
+        return _to_json_str([serialize_item(item) for item in value])
+
+
+def _to_json_str(obj: Any, indent: int | None = None) -> str:
+    return json.dumps(obj, ensure_ascii=False, indent=indent)
diff --git a/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_host/_elicitation.py b/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_host/_elicitation.py
index 373f995c231d..e5eda7cfd9c1 100644
--- a/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_host/_elicitation.py
+++ b/python/packages/autogen-ext/src/autogen_ext/tools/mcp/_host/_elicitation.py
@@ -12,6 +12,8 @@
 
 from mcp import types as mcp_types
 
+from .._base import _to_json_str
+
 _ELICITATION_CHOICE_SHORTHANDS = {"a": "accept", "d": "decline", "c": "cancel"}
 
 
@@ -82,7 +84,7 @@ async def elicit(self, params: mcp_types.ElicitRequestParams) -> mcp_types.Elici
                 prompt = "\n".join(
                     [
                         "Input Schema:",
-                        json.dumps(params.requestedSchema, indent=2),
+                        _to_json_str(params.requestedSchema, indent=2),
                         "Please enter a JSON string following the above schema: ",
                     ]
                 )
diff --git a/python/packages/autogen-ext/tests/tools/test_mcp_serialization.py b/python/packages/autogen-ext/tests/tools/test_mcp_serialization.py
new file mode 100644
index 000000000000..749bf691a39c
--- /dev/null
+++ b/python/packages/autogen-ext/tests/tools/test_mcp_serialization.py
@@ -0,0 +1,34 @@
+import pytest
+from autogen_ext.tools.mcp._base import _to_json_str
+from autogen_ext.tools.mcp._stdio import StdioMcpToolAdapter
+from mcp.types import TextContent, Tool
+
+
+def test_to_json_str_no_escape_japanese() -> None:
+    """Verify that Japanese text is not escaped in _to_json_str."""
+    data = {"text": "日本語"}
+    serialized = _to_json_str(data)
+
+    # Check that the literal Japanese characters are in the string
+    assert "日本語" in serialized
+    # Check that the escaped unicode sequence is NOT in the string
+    assert "\\u65e5\\u672c\\u8a9e" not in serialized
+
+
+def test_mcp_tool_adapter_return_value_no_escape_japanese() -> None:
+    """Verify that McpToolAdapter.return_value_as_string does not escape Japanese text."""
+    # Mock parameters
+    from autogen_ext.tools.mcp import StdioServerParams
+
+    server_params = StdioServerParams(command="echo", args=["test"])
+    tool = Tool(name="test_tool", description="A test tool", inputSchema={"type": "object", "properties": {}})
+
+    adapter = StdioMcpToolAdapter(server_params=server_params, tool=tool)
+
+    content = [TextContent(type="text", text="日本語")]
+    serialized = adapter.return_value_as_string(content)
+
+    # Check that the literal Japanese characters are in the string
+    assert "日本語" in serialized
+    # Check that the escaped unicode sequence is NOT in the string
+    assert "\\u65e5\\u672c\\u8a9e" not in serialized