Foundry Agent Issue: Fabric Data Agent Tool now working

Question

Issue with Microsoft Foundry Agent tool: Fabric Data Agent.
Getting error:

tool_user_error: Fabric run failed during execution. Find out more troubleshooting details here - https://aka.ms/foundryfabrictroubleshooting
I have checked the troubleshooting page but nothing useful as the error is "too generic"

Foundry Agent is using Model: Gpt-4.1
Tool connected: Fabric Data Agent

My Agent was working fine till few hours back and it's been working since last few days ever since it's been created.

I have admin access for my Entra ID in Fabric workspace and also Data Agent.
I have both Azure AI User and Azure AI Admin role at foundry service and project level.
Even provided the Foundry service project's Managed Identity Admin rights on the Fabric Workspace.
I have separately tested the Fabric Data Agent from Fabric workspace and it's working absolutely fine.
I have even created a new connection for the Fabric Data Agent and tried with a new agent, still it's same.
I have also tried by changing the LLM Model.

Not sure what happened suddenly! Can anyone help or suggest?
Also, is there any Quota limit?

Answer 1

Hello AyusmanBasu-6332,

I understand that you're running into a frustrating issue with the Fabric Data Agent while getting an error that says: "tool_user_error: Fabric run failed during execution." Since you indicated that you've already checked the troubleshooting page and it's not providing useful insights,

This strongly suggests the failure is occurring in the service-to-service execution layer between Foundry Agent and Fabric, not inside the model itself.

Below is a consolidated checklist covering configuration, permissions, networking, and quota angles.

1.Role & Identity Validation

Even if you’re Entra ID admin, what matters is the identity used by Foundry Agent at runtime.

Verify Foundry project Managed Identity has:

• • Azure AI User
  • Azure AI Project Manager
  • (If applicable) Cognitive Services OpenAI Contributor

The same Managed Identity has the required role in the Fabric Workspace.

Remove and re-add the Managed Identity to Fabric Workspace to force a fresh token trust evaluation.

Sometimes token audience validation or cached identity bindings desync internally.

2.Fabric Capacity / Concurrency Limits

This is a common cause.

Even if manual runs succeed in Fabric, external invocation from Foundry can fail if:

Fabric capacity is throttled

Concurrency limits are reached

Background job limits are exceeded

Capacity was paused or auto-scaled

Check Fabric capacity status, Workspace compute health, Active job count

There is no specific public “Agent Tool quota,” but Fabric capacity and cross-service throttling absolutely apply.

3.Resource & Network Configuration Changes

Since this stopped suddenly, check whether any changes occurred in:

VNET configuration

Private endpoints

DNS resolution

Firewall rules

Cross-region routing

If you are using private endpoints, confirm:

• The endpoint wasn’t modified after agent creation
• DNS still resolves correctly from Foundry service context

4.Regional Alignment

Confirm Foundry project region, Fabric workspace region

Cross-region invocation can introduce routing failures or latency-related execution failures.

Ideally, both services should be in the same region.

5.Agent & Connection Validation

Even though you recreated the connection, verify:

The Fabric Data Agent is published

The connection shows as healthy

The agent instructions clearly reference when to use the Fabric tool

The connection is properly selected in the Agent configuration

Sometimes a connection exists but isn’t fully published or bound.

6.Backend / Service Incident Possibility

Because it was working and then suddenly stopped, this is highly indicative of:

A backend update

A short-lived service regression

Token exchange failure between services

Check Azure Service Health for Azure AI Foundry, Microsoft Fabric

Even brief degradation can produce this exact generic tool_user_error.

7.Model Compatibility

Since changing models didn’t help, this is unlikely the root cause. However, confirm:

The model (e.g., GPT-4.1) is supported in your region

You are not hitting TPM/RPM model limits

If it were a model quota issue, you would typically see a quota or rate limit error rather than a Fabric execution failure.

Most Likely Causes

Given your scenario, the highest probability causes are:

1. Fabric capacity throttling
2. Managed Identity token trust refresh issue
3. Backend regression between Foundry and Fabric

Recommended Steps

Check Fabric capacity health and active job usage.

Remove and re-add Managed Identity permissions in Fabric.

Fully delete and recreate the Fabric connection.

Confirm region alignment.

Check Azure Service Health.

Given that it worked earlier without configuration changes, this is very unlikely to be a misconfiguration on your side and more likely a capacity or backend execution issue.

I Hope this helps. Do let me know if you have any further queries.

Thank you!

Answer 2

@AyusmanBasu-6332 try same query by switching to "Published mode" in Fabric Data Agent and see if the issue is with the Published version. That's what Foundry Agent connects to.

Answer 3

Hello AyusmanBasu-6332,

Thanks for raising it in the Q&A forum!

Authentication Mismatch: Fabric Data Agents only support user identity (OBO) authentication, not Service Principal Names (SPN) or managed identities. Even if the connection appears successful, data retrieval may silently fail if authentication isn't properly configured.

Permission Issues: The user identity accessing the Foundry agent must have read access to the Fabric workspace and underlying Lakehouse or semantic model. Workspace admin rights alone may not be sufficient—explicit read permissions on the data artifacts are required.

Tenant Mismatch: Your Fabric Data Agent and Foundry Agent must be in the same tenant. Cross-tenant scenarios are not supported.

Model Compatibility: Claude models have known compatibility issues with Fabric Data Agents. OpenAI-based agents (GPT-4, GPT-3.5) work properly, but Anthropic Claude models may fail to retrieve data even when the connection is established.

Service Principal Configuration: If you set up the Fabric Data Agent using a service principal, you may need to rotate the client secret and update it in Foundry. Expired or incorrect credentials cause authentication failures.

Resolution Steps

Verify Same Tenant Requirement:

Confirm both your Fabric workspace and Foundry project are in the same Azure AD tenant

Check tenant IDs in both portals to ensure alignment

Use Interactive User Authentication:

Sign into both Fabric and Azure AI Foundry with the same user account

Validate this user can directly query the data in the Fabric Data Agent playground

Recreate the Foundry connection from scratch using this user identity

Switch to OpenAI Models:
If you're using Claude models, switch to OpenAI models (GPT-4 or GPT-4o):

Navigate to your agent configuration in Foundry

Change the model deployment to an OpenAI-based model

Test the Fabric Data Agent connection again

Assign Foundry Service Principal Permissions:

Go to your Foundry project → Settings → Properties

Copy the Managed Identity or Service Principal Object ID

In Fabric workspace settings, add this service principal with Contributor permissions

Wait 10-15 minutes for permissions to propagate

Use Lakehouse Instead of Semantic Models:
If your Fabric Data Agent is connected to a semantic model, try migrating the data to a Lakehouse:

Semantic models have known issues with Foundry integration

Lakehouses provide more reliable connectivity

Check Region Alignment:
Ensure your Foundry project region and Fabric capacity region are in the same geography (e.g., both in East US or both in West Europe). Region mismatches can cause "Create assistant failed" errors.

Test in Fabric Data Agent Playground First:
Before troubleshooting Foundry integration:

Test your queries directly in the Fabric Data Agent playground

Confirm the agent returns expected results

If it fails in Fabric itself, the issue is with the Fabric agent configuration, not Foundry integration

Limitation: Only One Fabric Agent per Foundry Agent:
Currently, you can only add one Fabric Data Agent as a tool per Foundry agent. If you need multiple Fabric agents, consider creating separate Foundry agents or using an orchestrator pattern.

If this helps, kindly accept the answer.

Best Regards,
Jerald Felix