microsoft/semantic-kernel · error · FunctionExecutionException

Unsupported content type: {type(content)}

Error message

Unsupported content type: {type(content)}

What it means

Thrown by _kernel_content_to_mcp_content_types (mcp.py:184) as a FunctionExecutionException when the content argument is not one of TextContent, ImageContent, AudioContent, BinaryContent, or ChatMessageContent. The MCP connector can only serialize these kernel content types into MCP message types; anything else is rejected. Note: unsupported items inside a ChatMessageContent are silently skipped (debug-logged) rather than raising.

Source

Thrown at python/semantic_kernel/connectors/mcp.py:184

    if isinstance(content, BinaryContent):
        return [
            types.EmbeddedResource(
                type="resource",
                resource=types.BlobResourceContents(
                    blob=content.data_string, mimeType=content.mime_type, uri=content.uri or "sk://binary"
                ),
            )
        ]
    if isinstance(content, ChatMessageContent):
        messages: list[types.TextContent | types.ImageContent | types.AudioContent | types.EmbeddedResource] = []
        for item in content.items:
            if isinstance(item, (TextContent, ImageContent, BinaryContent, AudioContent)):
                messages.extend(_kernel_content_to_mcp_content_types(item))
            else:
                logger.debug("Unsupported content type: %s", type(item))
        return messages

    raise FunctionExecutionException(f"Unsupported content type: {type(content)}")


@experimental
def _get_parameter_dict_from_mcp_prompt(prompt: types.Prompt) -> list[dict[str, Any]]:
    """Creates a MCPFunction instance from a prompt."""
    # Check if 'properties' is missing or not a dictionary
    if not prompt.arguments:
        return []
    return [
        {
            "name": prompt_argument.name,
            "description": prompt_argument.description,
            "is_required": True,
            "type_object": str,
        }
        for prompt_argument in prompt.arguments
    ]

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Return one of the supported content types (TextContent/ImageContent/AudioContent/BinaryContent) or a ChatMessageContent composed of them.
2. If you have a custom type, convert it to TextContent(str(value)) before returning from the MCP-exposed tool.
3. Upgrade semantic-kernel: newer content types may have been added to the mapping.
4. For ChatMessageContent, ensure every item is one of the four supported leaf types; unsupported items are dropped silently.

Example fix

# before
return [FunctionResultContent(id=..., result=my_obj)]

# after
return [TextContent(text=str(my_obj))]

from semantic_kernel.contents import TextContent, ImageContent, AudioContent, BinaryContent, ChatMessageContent

SUPPORTED = (TextContent, ImageContent, AudioContent, BinaryContent, ChatMessageContent)

def contents_are_mcp_supported(value) -> bool:
    items = value if isinstance(value, list) else [value]
    return all(isinstance(i, SUPPORTED) for i in items)

from semantic_kernel.contents import TextContent, ImageContent, AudioContent, BinaryContent, ChatMessageContent

def is_supported_content(content) -> bool:
    return isinstance(content, (TextContent, ImageContent, AudioContent, BinaryContent, ChatMessageContent))

from semantic_kernel.exceptions.function_exceptions import FunctionExecutionException

try:
    result = await plugin.call_tool("foo")
except FunctionExecutionException as ex:
    if "Unsupported content type" in str(ex):
        # coerce the tool's return value to TextContent and retry
        ...

When it happens

Trigger: A kernel tool returns a content type the connector cannot map, e.g. a FunctionResultContent or FunctionCallContent passed at the top level, a custom KernelContent subclass, or a raw Python object wrapped incorrectly. The final raise at mcp.py:184 is reached only when none of the isinstance branches match.

Common situations: Returning a custom content subclass from a tool invoked over MCP; passing a FunctionResultContent directly instead of its inner value; version mismatch where a newer content type isn't handled; building tool results manually.

Related errors

• Failed to call tool '{tool_name}'.
• Failed to call prompt '{prompt_name}'.
• Plugin creation failed for {pluginName}
• Prompt name is required.
• No handler found for the prompt '{promptName}'.

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/c55bda86fff125a7. Report an issue: GitHub.