Python: Azure AI Foundry agent_id not exposed by BaseAgent

[Flerov]

Hi Team,
I encountered a problem when working with Azure AI Agents.
When an agent is created in Azure AI Foundry, it receives a unique agent_id (format: asst_...). However, this agent_id is not exposed anywhere on the BaseAgent (e.g. ChatAgent) API.
This makes it impossible to directly reference the created Foundry agent when using the AIProjectClient (azure.ai.projects.aio) unless the developer manually tracks the ID outside of the agent object.

Why this matters
Developers may need to interact with the same agent instance both through the agent_framework SDK (e.g. ChatAgent) and directly through the Foundry client (AIProjectClient). In general, since this is the agents primary key in the Azure AI Foundry, it is important that a developer know under which ID the agent got created. Without access to the agent_id, it is not possible to fetch, update, or manage the underlying Foundry agent.
It would be very helpful if the agent_id were exposed, for example via agent.metadata.agent_id or a similar property.

import asyncio
from agent_framework import ChatAgent
from agent_framework.azure import AzureAIAgentClient
from azure.ai.projects.aio import AIProjectClient
from azure.identity.aio import AzureCliCredential

PROJECT_ENDPOINT = "<YOUR-PROJECT-ENDPOINT"

async def main():
    async with (
        AzureCliCredential() as credential,
        AIProjectClient(
            endpoint=PROJECT_ENDPOINT ,
            credential=credential
        ) as project_client,
    ):
        agent1 = ChatAgent(
            chat_client=AzureAIAgentClient(
                project_client=project_client,
            ),
            name="BugAgent3"
        )
        await agent1.run("Hello!")

        # Currently this fails because `agent_id` is not exposed on BaseAgent
        await project_client.agents.get_agent(agent_id=agent1.metadata.agent_id)

        input("IDLE ")

if __name__ == '__main__':
    asyncio.run(main())

Expected behavior
The BaseAgent (e.g. ChatAgent) should expose the Azure AI Foundry agent_id.
Example: agent1.metadata.agent_id returns asst_xxxxx.

Actual behavior
The agent_id exists internally but is not exposed, so the developer cannot use it to interact with the Foundry API.

[markwallace-microsoft]

The AgentProtocol has an id property, see: https://github.com/microsoft/agent-framework/blob/main/python/packages/core/agent_framework/_agents.py#L164

[Flerov]

The AgentProtocol has an id property, see: https://github.com/microsoft/agent-framework/blob/main/python/packages/core/agent_framework/_agents.py#L164

This is a different ID then the one needed in the Azure AI Foundry

import asyncio
from agent_framework import ChatAgent
from agent_framework.azure import AzureAIAgentClient
from azure.ai.projects.aio import AIProjectClient
from azure.identity.aio import AzureCliCredential

PROJECT_ENDPOINT = "<YOUR-ENDPOINT>"

async def main():
    async with (
        AzureCliCredential() as credential,
        AIProjectClient(
            endpoint=PROJECT_ENDPOINT ,
            credential=credential
        ) as project_client,
    ):
        agent1 = ChatAgent(
            chat_client=AzureAIAgentClient(
                project_client=project_client,
                model_deployment_name='gpt-4.1-nano'
            ),
            name="BugAgent3"
        )
        await agent1.run("Hello!")

        # Currently this fails because `agent_id` is not exposed on BaseAgent
        await project_client.agents.get_agent(agent_id=agent1.id)

        input("IDLE ")

if __name__ == '__main__':
    asyncio.run(main())

[thiagofonsecabarros]

I was able to retrieve the Agent ID only through this listAgents.ts file

import dotenv from "dotenv";
import { AIProjectClient } from "@azure/ai-projects";
import { DefaultAzureCredential } from "@azure/identity";

dotenv.config();

async function main() {
const endpoint = process.env.AZURE_AI_PROJECT_ENDPOINT;
if (!endpoint) {
throw new Error("AZURE_AI_PROJECT_ENDPOINT not set");
}

const credential = new DefaultAzureCredential();
const projectClient = new AIProjectClient(endpoint, credential);

const agentsClient = projectClient.agents;

console.log("Listing agents in this project...");

// JS SDK: agentsClient.agents.listAgents() is the pattern in the docs
// but some versions expose just .listAgents(). We'll handle both.
const anyClient: any = agentsClient;
const listFn =
anyClient.agents?.listAgents?.bind(anyClient.agents) ??
anyClient.listAgents?.bind(anyClient);

if (!listFn) {
throw new Error(
"Could not find listAgents() on projectClient.agents; check SDK version."
);
}

const iterator = listFn();

for await (const agent of iterator) {
console.log(
Agent name: ${agent.name} | ID: ${agent.id} | Description: ${agent.description}
);
}
}

main().catch((err) => {
console.error("Error listing agents:", err);
process.exit(1);
});

[de-loots1994]

@markwallace-microsoft unfortunately this does not work, we have the same issue. Pretty tough for developers to keep up with changes if we can't keep up with how "id's" are named.

Now the method for id is simply the name, not what it use to be, in the old foundry UI experience. I can't find the ID anywhere and the SDK can not list the id in the previous format... gotta refactor again until the new update comes again </3

Becomes quite unworkable to keep up, where is the QA?

[Parbuster]

This issue still exists. In my code I use this:
agent = project_client.agents.get(agent_name="Bob2")
azClient = AzureAIAgentClient(credential=DefaultAzureCredential(), project_client=project_client, agent_id=agent.id)

But it crashes with Invalid 'assistant_id': 'Bob2'. Expected an ID that begins with 'asst'. And that is what the ids in the old foundry looked like.