Error The server had an error processing your request with A2A/MCP tool Foundry

Châu Lâm Đình Ái 0 Reputation points
2026-07-24T03:03:24.3033333+00:00

Describe the bug: We have an Azure Foundry instance in Japan East. We have noticed that my agents fail and return an error whenever they have tools connected. If we try prompting an agent that does not have any tools connected, it works perfectly fine without any issues. However, the moment the agent has connected tools and attempts to process a request, the server throws an error.

Expected behavior: The agent should successfully process requests and execute the connected tools without throwing a server error.

Actual behavior: Error The server had an error processing your request. Sorry about that! You can retry your request, or contact us through an Azure support request at: https://go.microsoft.com/fwlink/?linkid=2213926 if you keep seeing this error. (Please include the request ID 120b15c5-fd10-4abc-ab48-1cbb6c766128 in your email.) image

Reproduction Steps: Step 1. Open Azure Foundry in the Japan East region. Step 2. Test an agent with no tools connected (works normally). Step 3. Connect a agent which have at lease a tool. Step 4. Send a prompt to the agent. Step 5. The request fails with a server error.

Environment: This is experienced directly by the agent in the Azure Foundry Chat Playground.

Foundry Agent Service
Foundry Agent Service

A fully managed platform in Microsoft Foundry for hosting, scaling, and securing AI agents built with any supported framework or model

0 comments No comments

1 answer

Sort by: Most helpful
  1. AI answer

    2026-07-24T03:03:49.13+00:00

    This pattern matches either a tool discovery/connectivity problem or a transient service-side issue in Foundry, especially since agents without tools work and failures start only when tools are attached.

    Recommended checks:

    1. Validate tool discovery first.
      • If tools/list returns zero tools for MCP or A2A tools, the usual cause is invalid or missing connection credentials.
      • Verify that the project_connection_id exists in the Foundry project.
      • Verify the stored credentials are correct.
      • Try connecting to the MCP server directly to confirm authentication works.
      • If managed identity is used, verify the caller has the correct RBAC role assignments on the target resource.
    2. If the issue is with OpenAPI-based tools, validate the spec.
      • Confirm the OpenAPI document is valid 3.0 or 3.1.
      • Check that paths, operationId, and parameter schemas are valid.
    3. If fewer tools appear than expected, check allowed_tools.
      • Remove allowed_tools temporarily.
      • Call tools/list again.
      • Re-add only the exact tool names returned. Tool names are case-sensitive.
    4. If tools are built-in and return empty, retry after a short delay.
      • For built-in tools, an empty result can mean the toolbox is not fully provisioned yet.
      • Wait 10 seconds and retry.
    5. For MCP-specific failures, verify the endpoint and auth.
      • Confirm the MCP server URL is correct and reachable.
      • Check network connectivity and any firewall rules blocking outbound requests.
      • Confirm the selected authentication method is correct.
      • If the server uses an API key or OAuth token, verify the credential is valid and not expired.
      • If tools do not appear in the allowed tools list, refresh the tool list by closing and reopening the MCP tool dialog.
    6. If the tool is an Azure Language agent integration, verify common configuration issues.
      • For authentication errors, confirm the connection uses the Ocp-Apim-Subscription-Key header and that the value matches the Azure Language resource key.
      • If CLU or CQA is not used as expected, confirm those projects are deployed and connected to the correct resources.
      • If the MCP server returns 404 or connection errors, verify the endpoint URL uses the correct Foundry resource name and API version, and confirm the Azure Language resource is in a supported region.
    7. Consider a transient regional or portal-side issue.
      • A reported Foundry case in Japan showed agent failures such as agent load errors, 403 errors, and bad requests that recovered without configuration changes.
      • That behavior was attributed to a temporary service-side or portal-related issue, including backend disruption, authorization/cache desynchronization, metadata/indexing delay, or regional instability.

    Given the symptom that all tool-enabled agents fail in Japan East while tool-free agents work, the most direct path is:

    • test tools/list
    • verify each tool connection credential and project_connection_id
    • test the MCP endpoint directly
    • verify managed identity RBAC if used
    • retry after a short delay
    • if the issue clears without changes, treat it as a transient service-side issue

    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.