Formerly known as Azure AI Services or Azure Cognitive Services is a unified collection of prebuilt AI capabilities within the Microsoft Foundry platform
Hi Marco117 & thx for join me at Q&A platform,
that both Open WebUI and Postman are failing, which makes this less likely to be an Open WebUI-specific issue. But I wouldn't conclude there's a Foundry service outage just yet.
One thing stands out: you're using your Foundry project's API key as a Bearer token to authenticate to the Toolbox MCP endpoint.
Microsoft's current Toolbox documentation shows a different authentication pattern for direct MCP clients: a Microsoft Entra access token scoped to https://ai.azure.com/.default, sent in the Authorization: Bearer header.
That's an important distinction. A Foundry project API key isn't the same thing as an Entra OAuth access token, even if both can be used to authenticate to certain Foundry APIs.
I'd test the same Toolbox endpoint from Postman using a freshly acquired Entra token instead of the project API key. You can obtain one after signing in with Azure CLI:
az account get-access-token --resource https://ai.azure.com --query accessToken -o tsv
Make sure the signed-in identity has the appropriate Foundry project permissions, such as the Foundry User role. Use the exact Toolbox MCP endpoint, including ?api-version=v1.
If that works, we've narrowed the problem down to the authentication method rather than the Toolbox itself. It would also suggest that the previous API-key-based behavior shouldn't be relied on without confirmation from Microsoft.
If Entra authentication fails too, check the actual HTTP response status and headers. A 401 suggests token validation, expiration, or audience issues, while a 403 points more toward authorization. The generic Postman message doesn't tell us which one occurred.
Regarding whether something changed yesterday, I couldn't confirm a Microsoft-announced authentication change or service incident affecting Toolboxes on October 24. So I wouldn't claim this is a confirmed regression.
If the same configuration worked until October 24 and now fails across multiple clients, I'd raise a Foundry support case with the UTC timestamps, project region, Toolbox endpoint (without credentials), HTTP status, and request correlation IDs. Ask engineering whether the Toolbox gateway's authentication requirements or token validation behavior changed. Does your existing project API key still work for normal Foundry model inference requests, and have you tried connecting to the Toolbox using an Entra access token instead?
That comparison should tell us quite a lot.
rgds,
Alex
p.s. if my answer is help - pls mark it as an answer