How to use custom bearer token authentication in Azure AI Agent Service with OpenAPI Specified Tools

Question

How to use custom bearer token authentication in Azure AI Agent Service with OpenAPI Specified Tools

Jagan P 25

We are working on AzureA AI Agents with OpenAPI tools in Azure AI foundry.

I have configured an Open API tool with the json schema and Authentication method as (connection)

The Open API json schema is linked to a private webApplication (hosted in Kubernetes - https://forms-tstnf-internal.eu.hrx.alight.com) and the endpoints accepts only bearer token

"securitySchemes": {
      "Bearer": {
        "type": "apiKey",
        "description": "JWT Authorization header using the Bearer scheme. Example: \"Authorization: Bearer {token}\"",
        "name": "Authorization",
        "in": "header"
      }
    }

User's image

The open api agent is connected to the below connected resource.

User's image

I tried the agent in playground, but it is not working. Below is the error message

"tool_calls": [
          {
            "id": "call_ehoPT0t42vGgQiUKXBlnDzFS",
            "type": "openapi",
            "function": {
              "name": "FormPackage_GetAllPackages",
              "arguments": "{\"Status\":\"Open\"}",
              "output": "Error occured while trying to execute http request.",
              "error": "Aiohttp client error occurred",
              "curl": "curl -X GET 'https://forms-tstnf-internal.eu.hrx.alight.com/api/packages?Status=Open' -H 'Content-Type: application/json' -H 'Accept: */*' -H 'Authorization: REDACTED'"
            }
          }

I am not sure if the hardcoded bearer token is sent correctly in the request header

or if the private open api endpoint is not accessible from the azure agent.

I have attached the open API json schema and the Agent run steps logs.

Also, can you please let me know how to generate new bearer token for each request and add it to the request header before agent calls the open api endpoint.

Packages Agent runSteps.log

packages swagger.log

SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-05-21T17:24:52.44+00:00

Hi @Jagan P,

Following up to see if the below suggestion was helpful. And, if you have any further query do let us know.

Thank you!
SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-05-22T11:27:58.2366667+00:00

Hi @Jagan P,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

Thank you!
Jagan P 25 Reputation points

2025-06-04T15:22:48.1466667+00:00

Currently we are still working on exposing the API via Azure API Management (APIM)

so, I couldn't test it.

And I have a small doubt.
We are using Azure function app to invoke Azure Agents.

and the token which I need to use to call the API is short lived. I can generate the token each time before invoking the agent, but how to pass it to the Agent?

I mean how can I dynamically replace the Header Value Template: Bearer {{token}}
SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-06-09T10:28:17.48+00:00
Hi @Jagan P,

To use custom bearer token authentication with the Azure AI Agent Service and OpenAPI-specified tools, especially when dealing with short-lived tokens, the recommended approach is to dynamically inject the token at runtime instead of relying on static connections defined in Azure AI Studio.

Since you're invoking the agent from an Azure Function, you can generate the bearer token within your function and then pass it directly to the agent using the tools_override parameter, which is supported by the Azure AI Agent Service via the Assistants API (compatible with OpenAI).

This parameter allows you to override the tool’s authentication header per request. In your request payload, include the bearer token as part of the Authorization header like this: "Authorization": "Bearer <your-generated-token>".

The structure should resemble:

"tools_override": [ { "type": "openapi", "tool_id": "<YOUR_TOOL_ID>", "auth": { "type": "api_key", "key": "Authorization", "value": "Bearer <your-dynamic-token>" } } ]

Replace <YOUR_TOOL_ID> with the ID of the OpenAPI tool configured in Azure AI Studio and <your-dynamic-token> with the token generated by your function at runtime.

This method ensures that a fresh token is used on each invocation, which is essential for services requiring short-lived JWTs.

Avoid hardcoding the token or using static connections for dynamic authentication scenarios, as those approaches don’t support secure and seamless token rotation. For more control and security, this runtime injection via tools_override is the ideal solution.

Please refer this OAuth2 bearer authentication,

AAD bearer token.

I hope you understand.

Thank you!
Russell Myers 20 Reputation points

2025-06-11T06:12:47.55+00:00

Hello Srilakshmi,

To be clear, we are using an Azure AI Foundry Agent (AzureAIAgent) with an OpenAPI "Action" (tool) attached.
Can you confirm that the tools_override functionality you mention above is available for Azure AI Foundry agents (and if so, how is the tools_override parameter passed to this type of agent?)

Thanks,
Russell
(Note: I work with Jagan)
SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-06-23T12:31:57.4333333+00:00

Hi Russell Myers,

Apologies for the delayed response, and thank you for your patience.

To confirm, the tools_override parameter that allows dynamic runtime overrides such as injecting a fresh bearer token is currently not supported in Azure AI Foundry Agents when using OpenAPI tools (also referred to as Actions). While the tools_override feature exists in the OpenAI Assistants API, which Azure AI Studio is built on, the Azure AI Foundry interface (including Playground and function-triggered invocations) does not expose this functionality for direct use in agent configurations. As a result, dynamic per-request injection of headers such as Authorization: Bearer <token> is not currently possible through a built-in mechanism in Foundry.

Since your API requires a short-lived bearer token and you’re invoking the agent from an Azure Function, we recommend implementing a proxy pattern. In this approach, instead of calling the internal API directly from the AI agent, you would configure the OpenAPI tool to point to your Azure Function. The function would then generate the required token, make the authenticated call to your internal Kubernetes-hosted API, and return the result back to the agent. This gives you full control over authentication without relying on unsupported features. Alternatively, if you are working on exposing the API through Azure API Management (APIM), you can also configure APIM to inject the token using a policy or use managed identities, depending on how your backend is set up.

Thank you!

1 answer

Your answer

SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-05-21T17:24:52.44+00:00

Hi @Jagan P,

Following up to see if the below suggestion was helpful. And, if you have any further query do let us know.

Thank you!
SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-05-22T11:27:58.2366667+00:00

Hi @Jagan P,

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

Thank you!
Jagan P 25 Reputation points

2025-06-04T15:22:48.1466667+00:00

Currently we are still working on exposing the API via Azure API Management (APIM)

so, I couldn't test it.

And I have a small doubt.
We are using Azure function app to invoke Azure Agents.

and the token which I need to use to call the API is short lived. I can generate the token each time before invoking the agent, but how to pass it to the Agent?

I mean how can I dynamically replace the Header Value Template: Bearer {{token}}
SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-06-09T10:28:17.48+00:00

Hi @Jagan P,

To use custom bearer token authentication with the Azure AI Agent Service and OpenAPI-specified tools, especially when dealing with short-lived tokens, the recommended approach is to dynamically inject the token at runtime instead of relying on static connections defined in Azure AI Studio.

Since you're invoking the agent from an Azure Function, you can generate the bearer token within your function and then pass it directly to the agent using the tools_override parameter, which is supported by the Azure AI Agent Service via the Assistants API (compatible with OpenAI).

This parameter allows you to override the tool’s authentication header per request. In your request payload, include the bearer token as part of the Authorization header like this: "Authorization": "Bearer <your-generated-token>".

The structure should resemble:

"tools_override": [ { "type": "openapi", "tool_id": "<YOUR_TOOL_ID>", "auth": { "type": "api_key", "key": "Authorization", "value": "Bearer <your-dynamic-token>" } } ]

Replace <YOUR_TOOL_ID> with the ID of the OpenAPI tool configured in Azure AI Studio and <your-dynamic-token> with the token generated by your function at runtime.

This method ensures that a fresh token is used on each invocation, which is essential for services requiring short-lived JWTs.

Avoid hardcoding the token or using static connections for dynamic authentication scenarios, as those approaches don’t support secure and seamless token rotation. For more control and security, this runtime injection via tools_override is the ideal solution.

Please refer this OAuth2 bearer authentication,

AAD bearer token.

I hope you understand.

Thank you!
Russell Myers 20 Reputation points

2025-06-11T06:12:47.55+00:00

Hello Srilakshmi,

To be clear, we are using an Azure AI Foundry Agent (AzureAIAgent) with an OpenAPI "Action" (tool) attached.
Can you confirm that the tools_override functionality you mention above is available for Azure AI Foundry agents (and if so, how is the tools_override parameter passed to this type of agent?)

Thanks,
Russell
(Note: I work with Jagan)
SriLakshmi C 6,010 Reputation points Microsoft External Staff Moderator

2025-06-23T12:31:57.4333333+00:00

Hi Russell Myers,

Apologies for the delayed response, and thank you for your patience.

To confirm, the tools_override parameter that allows dynamic runtime overrides such as injecting a fresh bearer token is currently not supported in Azure AI Foundry Agents when using OpenAPI tools (also referred to as Actions). While the tools_override feature exists in the OpenAI Assistants API, which Azure AI Studio is built on, the Azure AI Foundry interface (including Playground and function-triggered invocations) does not expose this functionality for direct use in agent configurations. As a result, dynamic per-request injection of headers such as Authorization: Bearer <token> is not currently possible through a built-in mechanism in Foundry.

Since your API requires a short-lived bearer token and you’re invoking the agent from an Azure Function, we recommend implementing a proxy pattern. In this approach, instead of calling the internal API directly from the AI agent, you would configure the OpenAPI tool to point to your Azure Function. The function would then generate the required token, make the authenticated call to your internal Kubernetes-hosted API, and return the result back to the agent. This gives you full control over authentication without relying on unsupported features. Alternatively, if you are working on exposing the API through Azure API Management (APIM), you can also configure APIM to inject the token using a policy or use managed identities, depending on how your backend is set up.

Thank you!

Answer 1

Dear Jagan,

To resolve your issue with Azure AI Agent Service using OpenAPI tools and bearer token authentication, you’ll need to ensure two major things are handled properly:

1. Network Accessibility from Azure AI Agent Service

Azure AI Agent Service must be able to reach the internal Kubernetes service endpoint:

https://forms-tstnf-internal.eu.hrx.alight.com

Solution Options

:

Use Azure VNet integration: Your agent must be deployed in a VNet-enabled environment (e.g., Azure Container Apps with VNet, or Azure Functions Premium Plan with VNet support).
Private Endpoint or Internal DNS resolution must be available from the AI Agent’s execution environment.
Consider exposing the API via Azure API Management (APIM) with VNet + Private Link to securely bridge public Azure AI services and internal Kubernetes-hosted APIs.

2. Dynamic Bearer Token Injection in Azure AI Agent Tool

The agent won’t automatically rotate or inject dynamic bearer tokens unless you use a proper Connection setup and secure Azure Managed Identity or custom connection logic.

Since you mentioned:

“Authentication method as (connection)”

You should follow this pattern using “connection” as a security method, not hardcoding bearer tokens directly in OpenAPI schema.

Steps to Use Dynamic Bearer Token with Connection Authentication

A. Modify OpenAPI Security Scheme (keep this)

"securitySchemes": {

"BearerAuth": {

"type": "apiKey",

"name": "Authorization",

"in": "header",

"description": "JWT Bearer token. Format: Bearer {token}"

}

B. Add Security Requirement

"security": [

{

"BearerAuth": []

}

]

C. Use Azure AI Agent Service’s Connection Feature

In Azure AI Foundry, go to your tool → Connections tab
Select “Add Connection” → Choose type “API Key”
Set:
- Header Name: Authorization
  - Header Value Template: Bearer {{token}}
  - - Replace {{token}} with a key vault reference or a logic app URL that returns a token dynamically

D. Token Generation Options

If your API requires a short-lived JWT:

Option 1: Use Azure Logic App or Azure Function to generate and return a fresh token.
Option 2: Manually paste token in connection for now if automation is not in place.

E. Testing the Connection

In Azure AI Foundry Playground, test your agent with:

Network access working (step 1 solved)
Connection set with Authorization header using the live token

If this fails, the curl error should show the actual root cause:

403 → token issue
404 → wrong endpoint
timeout or aiohttp client error → VNet/firewall issue

Summary Checklist

Task	Status
Task	Status
Private API reachable from Azure AI Foundry?	Ensure via VNet, APIM, or Private Link
OpenAPI securitySchemes defined correctly?	Yes
Connection used for Authorization: Bearer {{token}}?	Must do
Token rotation setup?	Optional but ideal
Agent tool configured with the right connection?	Verify in Azure Portal

Would you like me to provide:

A sample Logic App that generates a bearer token dynamically?
A diagram explaining how OpenAPI + Agent + connection works with VNet?

Let me know! To resolve your issue with Azure AI Agent Service using OpenAPI tools and bearer token authentication, you’ll need to ensure two major things are handled properly:

1. Network Accessibility from Azure AI Agent Service

Azure AI Agent Service must be able to reach the internal Kubernetes service endpoint:

https://forms-tstnf-internal.eu.hrx.alight.com

Solution Options

:

Use Azure VNet integration: Your agent must be deployed in a VNet-enabled environment (e.g., Azure Container Apps with VNet, or Azure Functions Premium Plan with VNet support).
Private Endpoint or Internal DNS resolution must be available from the AI Agent’s execution environment.
Consider exposing the API via Azure API Management (APIM) with VNet + Private Link to securely bridge public Azure AI services and internal Kubernetes-hosted APIs.

2. Dynamic Bearer Token Injection in Azure AI Agent Tool

The agent won’t automatically rotate or inject dynamic bearer tokens unless you use a proper Connection setup and secure Azure Managed Identity or custom connection logic.

Since you mentioned:

“Authentication method as (connection)”

You should follow this pattern using “connection” as a security method, not hardcoding bearer tokens directly in OpenAPI schema.

Steps to Use Dynamic Bearer Token with Connection Authentication

A. Modify OpenAPI Security Scheme (keep this)

"securitySchemes": {

"BearerAuth": {

"type": "apiKey",

"name": "Authorization",

"in": "header",

"description": "JWT Bearer token. Format: Bearer {token}"

}

B. Add Security Requirement

"security": [

{

"BearerAuth": []

}

]

C. Use Azure AI Agent Service’s Connection Feature

In Azure AI Foundry, go to your tool → Connections tab
Select “Add Connection” → Choose type “API Key”
Set:
- Header Name: Authorization
  - Header Value Template: Bearer {{token}}
  - - Replace {{token}} with a key vault reference or a logic app URL that returns a token dynamically

D. Token Generation Options

If your API requires a short-lived JWT:

Option 1: Use Azure Logic App or Azure Function to generate and return a fresh token.
Option 2: Manually paste token in connection for now if automation is not in place.

E. Testing the Connection

In Azure AI Foundry Playground, test your agent with:

Network access working (step 1 solved)
Connection set with Authorization header using the live token

If this fails, the curl error should show the actual root cause:

403 → token issue
404 → wrong endpoint
timeout or aiohttp client error → VNet/firewall issue

Summary Checklist

Task	Status
Private API reachable from Azure AI Foundry?	Ensure via VNet, APIM, or Private Link
OpenAPI securitySchemes defined correctly?	Yes
Connection used for Authorization: Bearer {{token}}?	Must do
Token rotation setup?	Optional but ideal
Agent tool configured with the right connection?	Verify in Azure Portal

Best Regards,

Jerald Felix

Share via

How to use custom bearer token authentication in Azure AI Agent Service with OpenAPI Specified Tools

1 answer

Your answer