Azure Foundry Agent Responses.

(0) Share

Report

Posted on by JH-05051851-0

I have created an agent to send end-user's questions to a Foundry Agent; this Foundry agent has it's own tools/knowledge source that it is using to produce answers. In Copilot Studio, the first question will be answered and all the follow-on questions will not produce a new response. When I check my Foundry Agent's trace, I can see the subsequent questions and the Foundry Agent's answers, but these responses for some reason do not return to Copilot Studio. Is there any documentation or anyone else running into this issue?

Screenshot 2026-0...

Your file is currently under scan for potential threats. Please wait while we review it for any viruses or malicious content.

Categories:

Model context protocol

I have the same question (0)

All responses (2)

Answers (0)

Sort by

Vish WR 3,422 on at

Like (0)

Report
Copy link

Link copied!

@JH-05051851-0

We can't see the images yet. You can add images directly instead of file upload, as file upload takes time to appear.

Was this reply helpful? Yes No
Suggested answer

AP-26031104-0 Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!
Thanks for the detailed description — the trace evidence is really helpful here.

The fact that your Foundry agent receives and processes all questions but only the first response returns to Copilot Studio strongly points to one of these causes:

Most likely cause — Thread/conversation ID handling:
Azure AI Foundry agents use a thread ID to maintain multi-turn context. If your Copilot Studio integration creates a new thread on the first call but does not persist and pass that same thread ID on subsequent turns, the Foundry agent may be responding to a different thread than the one Copilot Studio is listening on. Check that:

The thread ID returned on the first call is stored in a variable
Every subsequent call passes that same thread ID rather than creating a new one

Second likely cause — Run status polling:
Foundry agent responses are asynchronous. The pattern requires polling the run status until it reaches completed, then retrieving the response. If your integration only polls once or has a short timeout, subsequent calls may return before the run completes and silently drop the response.

Third possibility — Output variable not updating:
In Copilot Studio, if the response is stored in a variable that is only set during the first action execution and the action node is not re-evaluated on follow-up turns (e.g., it's outside the main conversation loop), subsequent responses won't surface. Make sure the action call is inside a loop or trigger that fires on every user message.

Suggested checks:

In your Power Automate flow or HTTP action, log the thread ID on each call — confirm it's the same across turns

Add explicit run status polling with a do-until loop checking for completed or failed

Confirm the Copilot Studio topic that calls the Foundry agent is triggered on every new user message, not just the first

Was this reply helpful? Yes No