Run workflows in Azure Logic Apps from assistants in Foundry Agent Service (classic)

1: Create a logic app resource

To create a logic app resource and workflow for function calling, follow these steps:

1. In the Azure portal, create a Consumption logic app resource by following these general steps. After you open the workflow designer, return to this section.

2. On the designer, add the Request trigger named When an HTTP request is received by following these general steps.

3. Select the Request trigger to open the information pane and follow these steps:

   1. On the information pane, change the trigger name to reflect the function's task, for example: Function - Get weather forecast for today

   2. Provide the following trigger information:

   Parameter	Description
   Description	Enter a useful description about the task that the workflow performs, for example: This trigger and workflow gets the weather forecast for today from MSN Weather.
   Request Body JSON Schema	Enter a JSON schema that specifies the expected inputs. If you don't have a schema, select Use sample payload to generate schema, provide sample input, and select Done.

   The following image shows an example trigger with a task-relevant name, description, and schema:

   

4. On the designer toolbar, select Save to save the workflow.

   When you save a Request trigger for the first time, the designer generates the HTTP URL for the trigger's REST endpoint.

   

5. Based on your business use case or logic, add one or more actions in this workflow.

   For example, to get today's weather forecast for the current location, add an action from the MSN Weather connector. This example uses the Get forecast for today action.

   

6. In the action information pane for the Get forecast for today action, follow these steps:

   1. Select inside the Location parameter. When the input options appear (lightning icon and function icon), select the dynamic content list option (lightning icon).

      The dynamic content list lets you select outputs from preceding operations in the workflow. In this case, you want the location value that passed through the trigger to the workflow.

   2. From the dynamic content list, under trigger section named Function - Get weather forecast for today, select Location, for example:

      

      The action now looks like the following example:

      

7. On the designer, add the Response action by following these general steps.

   The workflow returns a response to Foundry by using the Response action.

8. From the dynamic content list, optionally select any prior operation outputs to return to Foundry. Or, you can provide a JSON schema to return the output in a specific format.

   

9. On the designer toolbar, select Save to save the workflow.

   Your workflow is now ready to import into Foundry.

2: Import the logic app resource as a function

To bring a logic app resource and workflow into Foundry for function calling through the Assistants playground, follow these steps:

1. In the Foundry portal (classic), open your Foundry project.

2. On the project sidebar, select Playgrounds. Scroll down and select Try the Assistants playground.

3. On the Assistants playground page, open an existing assistant or create a new one.

4. In the Setup section, expand Tools. Next to Functions, select + Add function, for example:

   

5. On the Add a custom function trigger screen, select the Logic Apps tab, for example:

   

6. On the Add logic app function screen, from the logic apps list, select a logic app:

   Note

   The list shows only logic apps with workflows that meet the requirements for function calling.

   

7. When you finish, select Save.

   The portal returns you to the Assistants playground page. In the Functions section, the successfully imported logic app workflow now appears as a function that your assistant can call, based on the user prompt.

3: Test the logic app function

To confirm that the imported logic app function works as expected, follow these steps:

1. On the Assistants playground page, in the chat window, enter a query for today's weather forecast in Seattle, for example:

   

2. To confirm the logic app function call, review the logs or the workflow run history.

   The following example shows a sample log with the function call:

   

1: Create a project client

Create a client object to connect to your Foundry project.

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

# Create the project client
project_client = AIProjectClient(
    endpoint=os.environ["PROJECT_ENDPOINT"],
    credential=DefaultAzureCredential(),
)

2: Register the logic app

Provide the trigger name and details to register the logic app resource. To find the AzureLogicAppTool utility code, visit the full sample on GitHub.

from user_logic_apps import AzureLogicAppTool

# Logic app details
LOGIC_APP_NAME = "your_logic_app_name"
TRIGGER_NAME = "your_trigger_name"

# Register the logic app with the agent tool utility
subscription_id = os.environ["SUBSCRIPTION_ID"]
resource_group = os.environ["resource_group_name"]

logic_app_tool = AzureLogicAppTool(subscription_id, resource_group)
logic_app_tool.register_logic_app(LOGIC_APP_NAME, TRIGGER_NAME)
print(f"Registered logic app '{LOGIC_APP_NAME}' with trigger '{TRIGGER_NAME}'.")

3: Create an agent with the logic app tool

Create an agent and attach the logic app as a function tool.

from azure.ai.agents.models import ToolSet, FunctionTool
from user_functions import fetch_current_datetime
from user_logic_apps import create_send_email_function

# Create the logic app function for the agent toolset
send_email_func = create_send_email_function(
    logic_app_tool, LOGIC_APP_NAME
)

functions_to_use = {fetch_current_datetime, send_email_func}

# Build and assign the toolset
functions = FunctionTool(functions=functions_to_use)
toolset = ToolSet()
toolset.add(functions)

agent = project_client.agents.create_agent(
    model=os.environ["MODEL_DEPLOYMENT_NAME"],
    name="SendEmailAgent",
    instructions="You are a specialized agent for sending emails.",
    toolset=toolset,
)
print(f"Created agent, ID: {agent.id}")

4: Create a thread

Create a thread and add a user message to start the conversation.

RECIPIENT_EMAIL = "your_recipient@example.com"

# Create a thread for communication
thread = project_client.agents.threads.create()
print(f"Created thread, ID: {thread.id}")

# Create a message in the thread
message = project_client.agents.messages.create(
    thread_id=thread.id,
    role="user",
    content=f"Send an email to {RECIPIENT_EMAIL} with the current date and time.",
)
print(f"Created message, ID: {message.id}")

5: Run the agent and check the output

Create a run and confirm that the agent uses the logic app tool to complete the task.

# Run the agent on the thread
run = project_client.agents.runs.create_and_process(
    thread_id=thread.id, agent_id=agent.id
)
print(f"Run finished with status: {run.status}")

if run.status == "failed":
    print(f"Run failed: {run.last_error}")

# Fetch and display all messages
messages = project_client.agents.messages.list(thread_id=thread.id)
for msg in messages:
    if msg.text_messages:
        last_text = msg.text_messages[-1]
        print(f"{msg.role}: {last_text.text.value}")

6: Clean up resources

Delete the agent when you're done to clean up resources.

# Delete the agent
project_client.agents.delete_agent(agent.id)
print("Deleted agent")