Azure AI Foundry Agent, Fabric Data Agent tool call gets stuck after first failure, subsequent turns show "No tool output found", even for non Fabric messages

Product

Azure AI Foundry Agents, Microsoft Fabric Data Agent tool integration (Preview)

Summary

I am connecting an Azure AI Foundry Agent to a Microsoft Fabric Data Agent that queries a Fabric Lakehouse. The Fabric Data Agent works in Fabric directly, and also works from one Foundry agent. However, in another Foundry agent within the same Foundry project and same region, the first Fabric tool call fails with "Fabric run failed during execution". After that first failure, the entire conversation becomes broken and every subsequent user message, even "hi", fails with "No tool output found for remote function call ...". It looks like the Foundry conversation gets into a bad state due to a missing tool output event.

This feels like a bug in the Foundry tool runtime, or a preview reliability issue, because it is stateful and cascades after the first failure.

Environment

1, MS AI Foundry Agents Playground 

2, Same Foundry project, same region, same user identity (Entra user) 

3, Same Fabric Data Agent tool, backed by a Fabric Lakehouse 

4, Model shown in the failing agent UI is gpt-4o 

5, Integration is Fabric Data Agent tool, label remote_functions, name azure_fabric

What works

The same Fabric Data Agent, same user identity, same question works in another Foundry agent in the same project and region.

Working trace includes both tool call and tool output:

• remote_function_call: azure_fabric, status OK 
• remote_function_call_output: azure_fabric, status OK, documents returned, plus a Fabric run URL 
• assistant responds successfully with citation to Fabric run URL

Example working response excerpt:

• user input, "what customers are there"
• agent output lists customers
• remote_function_call_output contains "documents" with Fabric response and a Fabric URL under /aiskills/.../stage/published/...

What fails

In the failing agent, the tool call step shows completed, but the overall response returns an error:

"tool_user_error: Fabric run failed during execution. Find out more troubleshooting details here https://aka.ms/foundryfabrictroubleshooting"

Then, after the first Fabric question, the chat breaks and subsequent prompts that should not require Fabric, for example "hi", "how are you", return:

"No tool output found for remote function call call_<id>"

This looks like Foundry is missing remote_function_call_output, then the conversation cannot recover.

Expected behavior

1, If the Fabric tool call fails, I expect the agent to fail only that turn, then the next turn should still work, at least for non tool messages 

2, Foundry should either surface the real Fabric error details, or allow the conversation to continue cleanly. Fabric Data Agent should not fail in Foundry.

Actual behavior

1, First Fabric tool call fails with a generic Foundry error 

2, Tool output event seems missing, no remote_function_call_output in traces 

3, Subsequent turns show "No tool output found for remote function call call_<id>" even for simple greetings 

4, The conversation becomes unusable unless I restart or use another agent

Evidence, failing trace excerpt

I see a RunStep like:

• name, remote_function_call: azure_fabric
• status, completed
• arguments, {"query":"List all customers with their IDs and names"}

Then the Response object ends with:

• status_code, ERROR
• description, tool_user_error: Fabric run failed during execution. Find out more troubleshooting details here https://aka.ms/foundryfabrictroubleshooting

After that, the UI shows:

• "No tool output found for remote function call call_<id>"

Evidence, working trace excerpt

The working agent shows:

• remote_function_call: azure_fabric, OK
• remote_function_call_output: azure_fabric, OK, output.documents present
• assistant uses returned content and provides a URL citation to the Fabric run

What I already checked and ruled out

1, Not a Fabric Lakehouse permission issue, because the same user and same Fabric Data Agent works from another Foundry agent in the same project and region. Not capacity issue, new Fabric capacity created.

2, Same Foundry project, same region, same tool selection, and I attempted to keep instructions identical 

3, This looks stateful, it only "breaks everything after" once a Fabric question is asked and fails, then the tool output is missing and subsequent turns fail

Questions

1, Is there a known issue where a missing Fabric tool output event causes the Foundry conversation to get stuck in the playground and when called from MS Agent Framework

2, Is this a known limitation of the Fabric Data Agent tool integration in Foundry preview, especially around tool output retrieval and correlation