Private Agentic Retrieval : Host name could not be resolved for Azure Search ( Foundry IQ ) knowledge base MCP.

[KalaivaniD12]

I have a private agentic retrieval configured. Followed this tutorial : https://learn.microsoft.com/en-us/azure/foundry/agents/how-to/foundry-iq-tutorial-private-overview. Getting this error, in the Foundry Portal, as well while the agent is being called programmatically.

error: Error Code: 400 , error encountered while enumerating tools from remote server: http://mysearchserv.search.windows.net:443/knowledgebases/supplier-knowledgebases-v1/mcp?api-version=****. Details: Name or Service not known the host name could not be resolved from the selected network path. Verify the server url and dns record from the intended Foundry network path. If the endpoint is private , internal or uses a custom domain, verify the capability host, Vnet, DataProxy, private DNS zone, gateway, proxy and firewall config so this service uses the correct private path. type: invalid request error, cod: tool user error

[Jannik Reinhard (JayRHa)]

The reported failure occurs during hostname resolution, before the request can reach authentication or knowledge-base validation. I would separate this into three checks:

1. If the URL in the error is the literal configured value, check the http://...:443 scheme/port combination. The configured Search endpoint should use HTTPS. This is separate from the DNS failure and may not explain that failure by itself.
2. Verify that the Search hostname resolves to the intended private address from the network path used by the agent runtime. Successful resolution from a laptop or another VNet does not establish that the managed runtime uses the same DNS path. Check private-endpoint approval and the relevant private DNS links or forwarding configuration.
3. Compare an in-VNet client call with the agent call using the same knowledge base and endpoint. If the client works and the agent fails, focus on the Foundry runtime's outbound/private connectivity configuration.

The private retrieval tutorial separates inbound connectivity, outbound dependencies, and end-to-end retrieval validation. Sharing the failing stage, runtime configuration and a redacted request/correlation ID should make this easier to isolate.