fix: handle Image helpers in mixed return annotations

src/mcp/server/mcpserver/utilities/func_metadata.py

@@ -340,6 +340,9 @@ def _try_create_model_and_schema(
     model = None
     wrap_output = False
 
+    if _contains_content_helper_type(type_expr):
+        return None, None, False
+
     # First handle special case: None
     if type_expr is None:
         model = _create_wrapped_model(func_name, original_annotation)
@@ -423,6 +426,21 @@ def _try_create_model_and_schema(
     return None, None, False
 
 
+def _contains_content_helper_type(annotation: Any) -> bool:
+    """Return whether an annotation contains an MCPServer content helper type."""
+    if isinstance(annotation, type) and issubclass(annotation, Image | Audio):
+        return True
+
+    args = get_args(annotation)
+    if not args:
+        return False
+
+    if get_origin(annotation) is Annotated:
+        return _contains_content_helper_type(args[0])
+
+    return any(_contains_content_helper_type(arg) for arg in args)
+
+
 _no_default = object()
 
 

tests/server/mcpserver/test_func_metadata.py

@@ -14,6 +14,7 @@
 
 from mcp.server.mcpserver.exceptions import InvalidSignature
 from mcp.server.mcpserver.utilities.func_metadata import func_metadata
+from mcp.server.mcpserver.utilities.types import Audio, Image
 from mcp.types import CallToolResult
 
 
@@ -716,6 +717,33 @@ def func_optional() -> str | None:  # pragma: no cover
     }
 
 
+def test_unstructured_output_content_helper_annotations():
+    """Image/Audio helper return annotations use content conversion, not schemas."""
+
+    def func_image() -> Image:  # pragma: no cover
+        return Image(data=b"abc", format="png")
+
+    def func_image_list() -> list[Image]:  # pragma: no cover
+        return [Image(data=b"abc", format="png")]
+
+    def func_nested_helpers() -> tuple[str, list[Image | Audio]]:  # pragma: no cover
+        return ("media", [Image(data=b"abc", format="png"), Audio(data=b"def", format="wav")])
+
+    def func_annotated_helper() -> Annotated[tuple[str, Image], "media"]:  # pragma: no cover
+        return ("image", Image(data=b"abc", format="png"))
+
+    for func in (
+        func_image,
+        func_image_list,
+        func_nested_helpers,
+        func_annotated_helper,
+    ):
+        meta = func_metadata(func)
+        assert meta.output_schema is None
+        assert meta.output_model is None
+        assert meta.wrap_output is False
+
+
 def test_structured_output_dataclass():
     """Test structured output with dataclass return types"""
 

tests/server/mcpserver/test_server.py

@@ -236,6 +236,36 @@ def mixed_content_tool_fn() -> list[ContentBlock]:
     ]
 
 
+def mixed_content_with_image_helper_tool_fn() -> tuple[str, Image, AudioContent]:
+    return (
+        "Hello",
+        Image(data=b"abc", format="png"),
+        AudioContent(type="audio", data="def", mime_type="audio/wav"),
+    )
+
+
+async def test_tool_mixed_content_with_image_helper_annotation():
+    mcp = MCPServer()
+    mcp.add_tool(mixed_content_with_image_helper_tool_fn)
+    async with Client(mcp) as client:
+        tools = await client.list_tools()
+        tool = next(tool for tool in tools.tools if tool.name == "mixed_content_with_image_helper_tool_fn")
+        assert tool.output_schema is None
+
+        result = await client.call_tool("mixed_content_with_image_helper_tool_fn", {})
+        assert len(result.content) == 3
+        content1, content2, content3 = result.content
+        assert isinstance(content1, TextContent)
+        assert content1.text == "Hello"
+        assert isinstance(content2, ImageContent)
+        assert content2.mime_type == "image/png"
+        assert content2.data == "YWJj"
+        assert isinstance(content3, AudioContent)
+        assert content3.mime_type == "audio/wav"
+        assert content3.data == "def"
+        assert result.structured_content is None
+
+
 class TestServerTools:
     async def test_add_tool(self):
         mcp = MCPServer()