Azure AI Search KB retrieve to Fabric IQ Ontology fails with Unsupported MCP method: notifications/initialized

I think Azure AI Search and Fabric IQ Ontology do not list notifications/initialized as a supported MCP call in the published preview specs.　

Hello @Sachitha Chithralekha

Thank you for reaching out to Microsoft Q&A.

Based on the information you've shared, it appears you've already performed thorough validation on the authentication flow and have successfully isolated the issue to the interaction between Azure AI Search / Foundry IQ and Microsoft Fabric IQ Ontology, rather than with delegated authentication or Azure AI Search itself.

From your testing, we understand that:

• Direct queries to the Fabric Eventhouse KQL endpoint using the delegated user token work successfully, and current_principal() resolves correctly, indicating that Row-Level Security (RLS) is functioning as expected.
• Azure AI Search accepts the delegated token (using the raw JWT in the x-ms-query-source-authorization header) and successfully reaches the Fabric IQ Ontology knowledge source.
• The request consistently fails only when Fabric IQ Ontology processes the request, returning:

    Unsupported MCP method: notifications/initialized
  

• The same behavior is reproduced using both:
  • Direct Azure AI Search KB Retrieve REST API
  • Azure.Search.Documents preview SDK (12.1.0-beta.1)

Since the issue reproduces across both REST and SDK implementations, it makes it less likely that the request format or SDK is the root cause.

Based on the current Microsoft documentation

The available documentation confirms that Fabric IQ Ontology is exposed through a Model Context Protocol (MCP) server (Preview) and supports a defined set of MCP tools for ontology operations. The documentation also notes that:

• Fabric IQ Ontology MCP is currently a Preview feature.
• Preview MCP capabilities, tool names, parameters, and behaviors may change before General Availability.
• The Fabric Ontology knowledge source is currently supported through the 2026-05-01-preview API version.

However, the current documentation does not describe or document an MCP method named:

notifications/initialized

nor does it specify whether Azure AI Search or Foundry IQ is expected to send this MCP notification during retrieval.

Because of this, we cannot confirm from the published documentation whether this method is expected behavior or whether Fabric IQ Ontology should currently support it.

Addressing your questions

1. Is Azure AI Search / Foundry IQ expected to send notifications/initialized to Fabric Ontology knowledge sources?

At this time, the published documentation does not specify whether Azure AI Search or Foundry IQ sends the notifications/initialized MCP method during retrieval. Therefore, we cannot confirm that this is expected behavior.

2. Is this a known compatibility issue between Azure AI Search Agentic Retrieval and Fabric IQ Ontology?

The documentation confirms that both Azure AI Search Agentic Retrieval and Fabric IQ Ontology MCP are Preview features, and preview MCP implementations are subject to change. While your observations are consistent with what could be a preview MCP contract mismatch, there is currently no published Microsoft documentation identifying this specific notifications/initialized error as a known issue.

3. Is there a required Fabric Ontology configuration or version flag?

Based on the current documentation, there is no documented configuration setting, feature flag, or version option that enables support for notifications/initialized. The published guidance only documents the supported MCP server endpoints and ontology tools.

Recommended validation steps

Although the error appears to occur downstream within the Fabric IQ Ontology MCP interaction, we recommend verifying the following items:

Confirm the Fabric Ontology knowledge source is using the supported 2026-05-01-preview API version.

Verify that the Azure AI Search knowledge source configuration, endpoint, authentication settings, and managed identity permissions are correctly configured.

Continue using the delegated JWT directly in the x-ms-query-source-authorization header, as your testing indicates this is the correct format for Fabric Ontology authentication.

Review Azure AI Search retrieval diagnostics and activity logs to confirm that the failure consistently occurs during the MCP call to the Fabric Ontology knowledge source and not earlier in the retrieval pipeline.

Based on the evidence you've provided:

• Delegated authentication is functioning correctly.
• Azure AI Search successfully forwards the request to the Fabric Ontology knowledge source.
• The failure occurs specifically when the Fabric IQ Ontology MCP endpoint receives the notifications/initialized method.

This makes the issue appear more consistent with a Preview MCP protocol compatibility issue between Azure AI Search Agentic Retrieval and the current Fabric IQ Ontology MCP implementation, rather than a problem with token propagation or request formatting.

Please refer this

Create a Fabric Ontology knowledge source (preview) (csharp) https://learn.microsoft.com/azure/search/agentic-knowledge-source-how-to-fabric-ontology

Fabric IQ Ontology MCP (preview) https://learn.microsoft.com/microsoft-copilot-studio/mcp-fabric-iq-ontology-work-iq

Connect a Foundry IQ knowledge base to Foundry Agent Service https://learn.microsoft.com/azure/foundry/agents/how-to/foundry-iq-connect

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

Thank you!