Azure AI Foundry agent tool POST fails with tool_server_error; GET tool works

Question

Azure AI Foundry agent tool POST fails with tool_server_error; GET tool works

Xavier Bailliet 0

We’re integrating Azure AI Foundry agent tools. A minimal GET tool works, but any POST tool fails before hitting our server.

What works

Tool: GET /api/ai-gateway/ping
Result: Success (pong)
Confirms network path from Azure to our endpoint works.

What fails

Tool: POST /api/ai-gateway/tool
Error: HTTP 500 (server_error: tool_server_error)
Request ID: a758f1bd-6c26-4fb8-8faa-e27ff0913780
UTC time: 2026‑01‑29 16:05:43
No server logs for the POST endpoint, so Azure fails before reaching our API.

What we tried

Simplified OpenAPI schema to minimal definitions
Removed unsupported const in schemas
Removed top-level security, used header params only
Added required headers (X-Correlation-Id, X-Organization-Context)
Published new agent versions
Confirmed gateway is reachable via GET tool

Question Why would POST tools consistently fail with tool_server_error while GET tools succeed, with no HTTP traffic reaching our server? Is there a known limitation or required schema field for POST tools in Azure AI Foundry agents?

**Minimal OpenAPI snippet (redacted server URL)**We’re integrating Azure AI Foundry agent tools. A minimal GET tool works, but any POST tool fails before hitting our server.

What works

Tool: GET /api/ai-gateway/ping
Result: Success (pong)
Confirms network path from Azure to our endpoint works.

What fails

Tool: POST /api/ai-gateway/tool
Error: HTTP 500 (server_error: tool_server_error)
Request ID: a758f1bd-6c26-4fb8-8faa-e27ff0913780
UTC time: 2026‑01‑29 16:05:43
No server logs for the POST endpoint, so Azure fails before reaching our API.

What we tried

Simplified OpenAPI schema to minimal definitions
Removed unsupported const in schemas
Removed top-level security, used header params only
Added required headers (X-Correlation-Id, X-Organization-Context)
Published new agent versions
Confirmed gateway is reachable via GET tool

Question
Why would POST tools consistently fail with tool_server_error while GET tools succeed, with no HTTP traffic reaching our server? Is there a known limitation or required schema field for POST tools in Azure AI Foundry agents?

Minimal OpenAPI snippet (redacted server URL)

{
  "openapi": "3.0.0",
  "info": { "title": "Validator Minimal", "version": "1.0.0" },
  "servers": [{ "url": "https://<redacted-host>" }],
  "paths": {
    "/api/ai-gateway/tool": {
      "post": {
        "operationId": "executeTool",
        "parameters": [
          { "name": "X-Correlation-Id", "in": "header", "required": true,
            "schema": { "type": "string", "format": "uuid" } }
        ],
        "requestBody": {
          "required": true,
          "content": {
            "application/json": {
              "schema": {
                "type": "object",
                "required": ["operation"],
                "properties": {
                  "operation": { "type": "string", "enum": ["validateModel"] },
                  "args": { "type": "object", "additionalProperties": true }
                }
              }
            }
          }
        },
        "responses": { "200": { "description": "OK" } }
      }
    }
  }
}

Anshika Varshney 15,625 Reputation points Microsoft External Staff Moderator

2026-01-29T17:55:34.5+00:00
Hi Xavier Bailliet,

Thanks for the detailed breakdown and for sharing the minimal repro that’s very helpful.

Based on the current Azure AI Foundry agent tools behavior, what you’re experiencing is a service‑side validation/gating issue, not a networking or server reachability problem on your end.

A few key clarifications from how agent tool invocation works today:

The fact that your GET tool succeeds confirms outbound connectivity, TLS, DNS, and routing from Foundry to your service are all working as expected.

The POST tool failing with tool_server_error before any request reaches your server indicates that the request is being rejected during Foundry’s internal tool validation/execution pipeline, before dispatch.

This failure happens prior to HTTP dispatch, which is why you see:

No inbound POST traffic

No API gateway logs

A generic HTTP 500 (server_error: tool_server_error) with a Request ID

From similar reports and internal behavior, we’ve seen, there are two important constraints to be aware of:

1. POST tools have stricter schema enforcement than GET tools

GET tools are currently more forgiving because they don’t include a request body. POST tools must pass Foundry’s internal OpenAPI validation, which is stricter than standard OpenAPI tooling.

Common causes for this exact symptom include:

Request body schemas that are too generic (additionalProperties: true at the root)

Enums combined with free‑form objects

Missing or ambiguous requestBody definitions (even if technically valid OpenAPI)

Properties that Foundry cannot deterministically materialize into tool arguments

Even when the schema validates externally, it can still fail Foundry’s internal tool compiler and result in an early tool_server_error.

2. The error is opaque by design (for now)

At the moment, when a POST tool fails during internal tool preparation, Foundry returns a generic error and does not surface:

The specific schema field that failed

A validation error message

A run‑step level diagnostic

This is why the Request ID is critical only the product team can inspect that execution path.

What you can do next

You’ve already done the right things by:

Minimizing the schema

Removing unsupported constructs

Republishing the agent

As a next step, we recommend raising a support request with Azure AI Foundry and including:

The Request ID

Timestamp (UTC)

Agent ID and project name

Confirmation that GET tools work but POST tools fail pre‑dispatch

Minimal OpenAPI example

There’s currently no documented functional limitation stating POST tools are unsupported, but there are known edge cases where POST schemas are rejected before execution.

Please feel free to update the thread if you hear back from support or need help framing the support request.

Thankyou!
Xavier Bailliet 0 Reputation points

2026-01-29T18:16:13.3466667+00:00

Hi Anshika — thanks for the detailed feedback, it’s very helpful.

One blocker on our side: even though we’re MS Partners, we still can’t get a support ticket created. The portal questions keep redirecting us back to Q&A, so we don’t have a path to submit a ticket with the Request ID.

If there’s a specific support category or intake route that avoids the Q&A loop for Azure AI Foundry tooling issues, please let us know. We can provide the Request ID, timestamps, and minimal OpenAPI repro immediately.

Thanks again.
Xavier Bailliet 0 Reputation points

2026-01-29T18:27:17.6233333+00:00

Hi Anshika — thanks for the detailed feedback, it’s very helpful.

One blocker on our side: even though we’re MS Partners, we still can’t get a support ticket created. The portal questions keep redirecting us back to Q&A, so we don’t have a path to submit a ticket with the Request ID.

If there’s a specific support category or intake route that avoids the Q&A loop for Azure AI Foundry tooling issues, please let us know. We can provide the Request ID, timestamps, and minimal OpenAPI repro immediately.

Thanks again.
Manas R Mohanty 17,270 Reputation points Moderator

2026-02-04T06:44:34.9066667+00:00

Hi Xavier Bailliet

I am trying to do a minimal replication with API gateway scenario with external model endpoints.

Could you please help with details requested in private message.

Thank you.
Manas R Mohanty 17,270 Reputation points Moderator

2026-02-12T10:35:11.06+00:00

Hi Xavier Bailliet

Good day. We could not hear from you.

Please help with details requested in private message if you are still facing internal server error issue or looking for further assistance.

Thank you for testing AI gateway Services.

Your answer

Xavier Bailliet 0 Reputation points

2026-01-29T18:16:13.3466667+00:00

Hi Anshika — thanks for the detailed feedback, it’s very helpful.

One blocker on our side: even though we’re MS Partners, we still can’t get a support ticket created. The portal questions keep redirecting us back to Q&A, so we don’t have a path to submit a ticket with the Request ID.

If there’s a specific support category or intake route that avoids the Q&A loop for Azure AI Foundry tooling issues, please let us know. We can provide the Request ID, timestamps, and minimal OpenAPI repro immediately.

Thanks again.
Xavier Bailliet 0 Reputation points

2026-01-29T18:27:17.6233333+00:00

Hi Anshika — thanks for the detailed feedback, it’s very helpful.

One blocker on our side: even though we’re MS Partners, we still can’t get a support ticket created. The portal questions keep redirecting us back to Q&A, so we don’t have a path to submit a ticket with the Request ID.

If there’s a specific support category or intake route that avoids the Q&A loop for Azure AI Foundry tooling issues, please let us know. We can provide the Request ID, timestamps, and minimal OpenAPI repro immediately.

Thanks again.
Manas R Mohanty 17,270 Reputation points Moderator

2026-02-04T06:44:34.9066667+00:00

Hi Xavier Bailliet

I am trying to do a minimal replication with API gateway scenario with external model endpoints.

Could you please help with details requested in private message.

Thank you.
Manas R Mohanty 17,270 Reputation points Moderator

2026-02-12T10:35:11.06+00:00

Hi Xavier Bailliet

Good day. We could not hear from you.

Please help with details requested in private message if you are still facing internal server error issue or looking for further assistance.

Thank you for testing AI gateway Services.