Automate workflows in Azure SRE Agent

Your team probably has recurring tasks like checking service health every morning, reviewing overnight alerts, verifying certificate expirations, or posting weekly capacity reports. Connect your tools, build a workflow, and let the agent run it on a schedule.

What you accomplish

Connect a notification tool so the agent can send messages
Create a custom agent that uses that tool
Schedule a recurring health check that runs automatically
See the complete workflow on the Agent Canvas

Prerequisites

Requirement	Details
Completed articles	Create and set up and Team onboarding.
Role on the agent	The SRE Agent Administrator role covers every step in this article. Other roles cover only part of the work. An Author can create the custom agent but not the scheduled task, and a Standard User can create the scheduled task but not the custom agent. Adding a connector requires the Administrator role. See User roles and permissions.
An account to send from	A Microsoft 365 account with an active Outlook mailbox. The agent signs in as this account, and the reports it sends arrive from this address.
Consent to authorize the connector	If your organization restricts which applications people can approve, an administrator might need to consent to the sign-in before the connector connects.
Azure context for this example	Access to the Azure resources the task checks. A logging provider is needed only if the task requests logs. The agent's identity needs suitable Azure role-based access control (RBAC) permissions for the requested data. See Agent permissions.
Recipient	An email address that can receive the test report.

Tip

Run your first investigation and Automate incident response aren't required for this article, but completing them first gives you a better understanding of what workflows can automate.

How it works

Automation connects three building blocks, each one set up through the portal:

Building block	What it does	Example
Connector	Gives the agent access to an external service	Outlook, Teams, Datadog, Splunk
Custom agent	A specialized worker with access to specific tools	`health-check-reporter` with permission to send messages
Scheduled task	Triggers a custom agent on a recurring schedule	"Every morning at 8 AM, check resource health and send a summary"

In the portal, the second block appears as Custom Agent.

The following walkthrough builds a morning health report delivered by email, but the same three blocks cover other recurring work. Change the instructions and the schedule to review overnight alerts, check certificate expirations, or post a weekly capacity summary. A weekly report sent to a distribution list also gives your leadership a standing record that nobody has to assemble by hand.

Add a connector

Connectors let the agent interact with external services. Start with a notification tool so your agent can report findings. These steps set up an Outlook connector. For Teams, see Set up a Teams connector.

Two identities are in scope before you begin. Your sign-in authorizes the agent to act in Outlook, which is why the mail it sends comes from your account. The agent uses its own user-assigned managed identity to reach Azure and gather the resource data for the report, so the RBAC roles assigned to that identity decide what the report can include. Azure creates that identity with the agent, so you don't create one here. For more information, see Agent identity and Agent permissions.

Go to Build + setup > Connectors in the side navigation.
Select Add connector.
Select the Notification tab.
Select the Office 365 Outlook connector for the Microsoft Outlook service.
Select Next.
Sign in and authorize access, and then select Next.
In Set up tools, select the email-send operation. The wizard rail labels this step Configure tools.
Set its Permission to Ask.
Turn on Parameter policy for the email-send operation.
In its recipient or To parameter, enter the recipient address from the prerequisites.
Select Next.
Review the configuration, and then select Create.

Checkpoint: The connector appears in the connectors list with a status of Connected.

Important

Ask pauses for your approval during a chat, but a scheduled task running in autonomous mode calls the operation without asking. Locking the recipient address is what keeps an unattended run from mailing the wrong person. For more information, see Run modes.

Tip

You can also add connectors based on the Model Context Protocol (MCP) for Datadog, Splunk, Elasticsearch, Dynatrace, New Relic, and custom MCP servers. See MCP connectors for the full list.

Create a custom agent

Custom agents are specialized workers with access to specific tools. You build and connect them on the Agent Canvas, a visual diagram of your custom agents, their tools, and the triggers that start them.

Go to Build + setup > Agent Canvas in the side navigation.
Select Create in the toolbar, and then select Custom Agent. The creation dialog opens with two tabs, Form and YAML.
Fill in the required fields:
- Custom agent name: for example, health-check-reporter
- Instructions: describe what this custom agent does and include the report recipient, for example, "You are a health check reporter. Check Azure resource health for my container apps and email the summary to ops@example.com."
Select Choose tools, and then select the email-send tool from your connector along with the Azure tools the report needs. Tools are the individual operations the agent can call. Choosing tools here overrides the global tool defaults for this custom agent, and leaving the selection empty lets it inherit all global tools. See Tools and Tools and skills.
Select Create.

Good instructions name the scope, the check, and what to do with the result. The example instruction works because it says which resources to look at, what to look for, and where the summary goes, so the agent doesn't have to guess any of the three. The YAML tab shows the same definition as text, so you can review or copy what you built.

The custom agent appears as a node on the canvas, and its selected Outlook operation shows up in a toolbox node, typically with an MCP badge.

Checkpoint: The canvas shows the custom-agent node and a toolbox node containing the selected Outlook tool.

Tip

The custom agent form also lets you assign skills and hooks. Skills are reusable procedures the agent loads when they're relevant, and hooks are controls that run before or after an action to keep it within bounds. See Create a subagent for the full walkthrough, Skills, and Agent hooks.

Schedule a recurring task

Link a scheduled task to the custom agent so it runs automatically.

Select the + button on the side of your custom-agent node.
Under the Trigger group, select Add scheduled task. The dialog opens with Response custom agent already set to that agent.

Fill in the fields:

Field	Example value
Task name	`daily-health-report`
Response custom agent	Your custom agent, already selected
Task details	Check the health of the resources in my resource group. Verify all apps are running, check CPU and memory metrics over the last hour, and review any recent warning logs. Summarize findings and send the report.
Frequency	Daily
Time of day	8:00 AM

Select Create task.

The Time of day label shows the time zone your browser is in, such as Time of day (PST), but the schedule is saved as UTC without a time zone. The task always runs at the same UTC time, so after a daylight saving change it runs an hour earlier or later than the local time you picked. If you set Frequency to Custom cron, the time picker is replaced by a Cron expression (UTC) field, and the expression is read as UTC.

Every run consumes agent usage, and a daily task keeps consuming it until you turn the task off. Pick the lowest frequency that still gives you the signal you need. For how usage is measured, see Pricing and billing.

The canvas now shows the complete workflow chain.

Verify the outcome

Run the task immediately to verify everything works:

Go to Automation in the side navigation, where you can see your scheduled tasks. Older links to the scheduled tasks page redirect here.
Select the checkbox next to your task.
Select Run now in the toolbar.
To watch the run, select your task name to open the executions view, and then select the Thread name link. You can also find the thread under Chats in the sidebar.

Run now submits the execution request and refreshes the task list. What the run produces, from the health checks and metrics through to the emailed report, depends on the tools you selected, the agent's access to your resources and telemetry, and the health of the connector.

Checkpoint: An execution appears for the task. If the execution resolves a thread, the executions view shows a Thread name link. Otherwise it shows No thread found. If the execution succeeds and invokes the Outlook tool, verify that the recipient receives the report.

If the task doesn't run

A scheduled task can finish without producing the result you expected. Start with the execution history.

Go to Automation and select the task name. Each execution row shows the start time, a Success or Failed status, and either a Thread name link or No thread found. The row doesn't show an error message, so open the thread to see what happened.
Open the most recent thread. It shows how the agent planned the work, which tools it called, and what it concluded.
An execution is marked successful once its work is handed to a thread, so a run can show Success even if the agent later stopped, paused for approval, or hit a failed tool call. Read the thread before you trust the status.
If no execution appears at all, confirm the task status is On and check the next run time.

What you see	Likely cause	What to do
The execution shows Success, but no email arrives	The email operation isn't enabled on the connector, the recipient parameter is locked to a different address, or the agent never reached the send step	Open the thread to see whether the agent called the Outlook tool. Then edit the connector, confirm the send operation is selected, and check the locked To value
The execution shows Success, and the thread is waiting for approval	The task's Agent autonomy level is Review, so the agent pauses in the thread before it acts	Open the thread and approve the pending action, or edit the task and set Agent autonomy level to Autonomous
The execution shows No thread found	The run failed before it created a thread	Confirm the connector status is Connected and the task's Response custom agent still exists, then select Run now to try again
The connector isn't Connected	The authorization expired, or the account lost access to the mailbox	Edit the connector and sign in again
The report arrives with no resource data	The agent's identity doesn't have access to the resources the task checks	Review the assignments described in Agent permissions

For more troubleshooting and task management, see Create and edit scheduled tasks.

Clean up resources

After you verify the report, use Turn off or Delete for a scheduled task you don't plan to keep. Use Remove or Disconnect for the connector. Revoke the Outlook OAuth grant when it's no longer needed.

You completed the getting-started journey

You built a workflow end to end. A connector gives your agent a way to reach Outlook, a custom agent knows what to check and where to report it, and a scheduled task runs the whole chain on its own.

Your agent can now:

Automate workflows: Run scheduled tasks through custom agents and connectors.
Use team context: Draw on the knowledge sources and team details you connected during setup and onboarding.

If you completed Run your first investigation, you also saw the agent diagnose a problem using your code, telemetry, and Azure context. If you completed Automate incident response, it picks up alerts that match a response plan and investigates them without being asked.

Capability	What it adds
Scheduled tasks	Advanced scheduling options and task management
Connectors	How connectors provide tools to your agent
Custom agents	How custom agents delegate and specialize work

Now that you're set up, explore Concepts to understand how the agent works, or try a tutorial for hands-on guidance on advanced features.

Where to go	What you'll find
Concepts	How roles, permissions, memory, connectors, and agent reasoning work
Capabilities	Detailed pages on every feature the agent offers
Tutorials	Step-by-step guides for deep investigation, connectors, hooks, and more

Feedback

Was this page helpful?

Last updated on 2026-09-25