Hosted agent on foundry: Unable to send response to the paused conversation

[karun19]

I have a hosted agent on Foundry that implements a multi-agent workflow. As part of the workflow, there is a step that requires human input, during which the flow pauses and waits for a response.

I’m able to test this successfully in the foundry playground, where I can provide the required input after the pause and the workflow resumes as expected.

However, I’m facing an issue when trying to do this programmatically. When I send the response after the flow pauses, it is treated as a new input, and a new conversation is started, instead of resuming the existing workflow.

Below is my client code.

`from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient

PROJECT_ENDPOINT = "https://xxx.services.ai.azure.com/api/projects/xxx-aifoundry"
AGENT_NAME = "agent-with-mcp"
AGENT_VERSION = "35"

def main():
client = AIProjectClient(
endpoint=PROJECT_ENDPOINT,
credential=DefaultAzureCredential(),
)

openai_client = client.get_openai_client()

# 1. Trigger the flow
print("🚀 Starting agent flow...")

start_response = openai_client.responses.create(
    input=[
        {
            "role": "user",
            "content": "Investigate HRMS production outage and remediate."
        }
    ],
    extra_body={
        "agent_reference": {
            "name": AGENT_NAME,
            "version": AGENT_VERSION,
            "type": "agent_reference"
        }
    }
)

print("Response ID:", start_response.id)
print(start_response.output_text)

# 2. Detect HITL pause (FIXED)
paused = any(
    getattr(item, "type", None) in ("mcp_approval_request", "tool_call")
    for item in start_response.output
)

if not paused:
    print("✅ Flow completed without approval.")
    return

print("⏸️ Flow paused for approval.")

# 3. Send approval
approval_response = openai_client.responses.create(
    previous_response_id=start_response.id,
    input=[
        {
            "role": "user",
            "content": "APPROVED"
        }
    ],
    extra_body={
        "agent_reference": {
            "name": AGENT_NAME,
            "version": AGENT_VERSION,
            "type": "agent_reference"
        }
    }
)

print("✅ Flow resumed and completed.")
print(approval_response.output_text)

if name == "main":
main()`

[microsoft-github-policy-service]

Thank you for contacting us! Any issue or feedback from you is quite important to us. We will do our best to fully respond to your issue as soon as possible. Sometimes additional investigations may be needed, we will usually get back to you within 2 days by adding comments to this issue. Please stay tuned.

[a1exwang]

Hi @karun19, this seems to be a questions about how to implement human in the loop with azure-ai-projects SDK. The azure sdk repo is a better place for this kind of questions and they should know more about it. Could you open issue there? https://github.com/Azure/azure-sdk-for-python/issues.