Formerly known as Azure AI Services or Azure Cognitive Services is a unified collection of prebuilt AI capabilities within the Microsoft Foundry platform
Hello @Ernie Gunning
Welcome to Microsoft Q&A .Thank you for reaching out to us.
After reviewing the current Responses API documentation and event model, the behavior observed appears to be related to how MCP events are surfaced through the Responses API stream. The service currently provides visibility into MCP tool discovery and execution lifecycle events, but the documented event model does not currently include an event designed to expose intermediate progress notifications generated by a remote MCP server during tool execution
- Regarding does Azure AI Foundry Agent Service support forwarding remote MCP progress notifications through the Responses API stream -
Based on the currently documented Responses API event model, remote MCP progress or notification messages are not exposed through the Responses API stream. The available MCP-related events provide lifecycle visibility for tool execution, such as when a tool call starts, completes, or fails, but no documented event currently exists that carries detailed progress payloads emitted by the remote MCP server.
- Regarding If there is an event type or configuration that enables these notifications?
The currently documented MCP-related streaming events include tool discovery events, tool argument generation events, and tool execution lifecycle events.
For reference:
- response.mcp_call.in_progress indicates that tool execution has started.
- response.mcp_call.completed indicates successful completion.
- response.mcp_call.failed indicates execution failure.
- response.mcp_call_arguments.delta streams tool arguments as they are being generated and does not represent backend execution progress.
At this time, no documented SDK option, request parameter, or service configuration has been identified that enables passthrough of MCP progress notifications through the Responses API stream.
- For the recommended approach for production scenarios requiring progress updates-
For long-running operations that require detailed progress reporting, the recommended pattern is to separate operation progress from the Responses API stream.
Typical implementation approaches include:
- Returning a job ID or operation ID from the MCP tool.
- Tracking execution progress independently using a correlation identifier.
- Surfacing progress through a dedicated channel such as:
- Server-Sent Events (SSE)
- WebSocket or Azure SignalR Service
- Status polling endpoint
- Event or message-based notification services
This allows detailed progress reporting while the Responses API continues to provide execution lifecycle events.
- Recommended validation and troubleshooting steps are
- Capturing the raw text/event-stream output returned by the Responses API.
- Comparing the raw stream with SDK-processed events to rule out client-side filtering.
- Confirming that the remote MCP server is actively emitting progress notifications during execution.
- Validating that the MCP server receives and uses the required progress token when generating progress notifications.
- Recording the model version, API version, SDK version, region, request ID, and timestamps for correlation purposes.
The following references might be helpful , please check them out
- Microsoft Foundry REST Reference - Azure OpenAI - Responses | Microsoft Learn
- Use the Azure OpenAI Responses API - Microsoft Foundry | Microsoft Learn
Please let us know if the response was helpful
Thank you