MS Foundry Agent published successfully but returns "does not have any deployments associated with it" when called via HTTP

Question

I created an agent within Foundry using the new interface and published it to MS Teams and Copilot 365. I called the agent from a Power Automate flow with the HTTP action. URI is https://AIAgentsFoundryResource.services.ai.azure.com/api/projects/{Project}/applications/{agent}/protocols/openai/responses?api-version=2025-11-15-preview. Authenticated by first obtaining a token via a separate HTTP action. Testing the flow results in the error: Action 'HTTP_Call_Agent' failed: Application 'Test4' does not have any deployments associated with it. [Request ID: 193aeef5379c182ec402caa852549812].  I have older agents that run successfully.  I believe this is know issue.  Are there any workarounds or updates and when this will be resolved.

 

Thanks for your help!

Answer 1

The error indicates that the Agent Application resource exists, but there is no running deployment associated with it. This aligns with the message “Application 'Test4' does not have any deployments associated with it” when calling the Responses endpoint.

Use the following checks and steps:

1. Verify the Agent Application and deployment
   • Confirm that the agent version was actually published and that an Agent Application resource plus a deployment were created.
   • In the Foundry portal, ensure the agent version shows a published status.
   • If using management APIs, get the deployment resource and confirm it exists and has a state:

          GET https://management.azure.com/subscriptions/{subscription_id}/resourceGroups/{resource_group}/providers/Microsoft.CognitiveServices/accounts/{account_name}/projects/{project_name}/applications/{application_name}/agentdeployments/{deployment_name}?api-version={api_version}
     

2. Ensure the deployment is running
   • Prompt/workflow agent deployments typically start automatically, but if the deployment is stopped, the application will behave as if it has no active deployment.
   • If state indicates the deployment is stopped, start it:

          POST https://management.azure.com/subscriptions/{subscription_id}/resourceGroups/{resource_group}/providers/Microsoft.CognitiveServices/accounts/{account_name}/projects/{project_name}/applications/{application_name}/agentdeployments/{deployment_name}/start?api-version={api_version}
     

   • After starting, retry the HTTP call from Power Automate.
3. Verify publishing and endpoint invocation
   • Confirm publishing succeeded by checking:
     1. The Agent Application resource exists.
     2. The deployment is running.
     3. The application endpoint can be invoked successfully.
   • For a quick verification, use the Responses protocol directly (outside Power Automate) with an access token for https://ai.azure.com:

          az account get-access-token --resource https://ai.azure.com
          
          curl -X POST \
            "https://<foundry-resource-name>.services.ai.azure.com/api/projects/<project-name>/applications/<app-name>/protocols/openai/responses?api-version=2025-11-15-preview" \
            -H "Authorization: Bearer <access-token>" \
            -H "Content-Type: application/json" \
            -d '{"input":"Say hello"}'
     

   • If this call succeeds, reuse the same pattern (URI, token resource, and body) in the Power Automate HTTP action.
4. Check RBAC and token scope
   • Ensure the caller used in Power Automate has the Azure AI User role (or equivalent) on the Agent Application resource, as Responses calls require this role.
   • The access token must be requested for https://ai.azure.com; using another resource (for example, https://management.azure.com) will cause authorization issues.
5. Known limitations and behavior
   • Agent Applications only expose the stateless Responses API; other APIs such as /conversations or /containers are not available. Ensure the Power Automate HTTP action is calling the /protocols/openai/responses path as shown above.
   • Multi-turn conversations must be managed on the client side; each call is stateless.

If older agents work but newly created ones fail, compare:

• Whether the new Agent Application actually has a deployment and that its state is running.
• Whether the same token acquisition and RBAC setup are used.

There is no documented workaround beyond ensuring a valid, running deployment is associated with the Agent Application and that the endpoint is invoked with the correct token and path.

---

References:

• Publish your agent as an Agent Application