Add computer use standalone tools to agents and agent flows (preview)

[This article is prerelease documentation and is subject to change.]

The computer use standalone tool is a modular version of the local in-agent computer use tool (hosted by, and tied to an agent). You can copy it as a local tool to an agent, or add it as a node to an agent flow. By using this standalone tool, you can create, test, and govern AI-driven automations that interact with digital interfaces. After you publish the standalone computer use tool, you can reuse it across multiple agents and agent flows, providing a centralized way to manage complex UI-based tasks.

By using the computer use standalone tool, you can:

  • Create modular and reusable automation building blocks.
  • Execute intelligent UI interactions based on natural language instructions.
  • Centralize governance by defining allowed applications and websites.
  • Monitor execution with high-fidelity session replays and detailed activity logs.
  • Incorporate human-in-the-loop supervision for sensitive or complex tasks.

Important

This article contains Microsoft Copilot Studio preview documentation and is subject to change.

Preview features aren't meant for production use and may have restricted functionality. These features are available before an official release so that you can get early access and provide feedback.

If you're building a production-ready agent, see Microsoft Copilot Studio Overview.

Create a computer use standalone tool

Note

Computer use as a standalone tool is currently rolling out across supported regions and might not be available in your region yet.

  1. In Copilot Studio, select Tools on the side bar. The Tools page appears. Learn more in Use shared tools from the Tools page.

  2. Select New tool. The New tool panel appears.

  3. Select Computer use. The configuration page appears, with its Designer tab in focus.

  4. Switch to the Overview tab.

    1. Select Edit on the Details tile. The Edit computer use details panel appears.

    2. Enter a name and a description for your tool. These details help other makers understand when to use this tool in their flows.

    3. Select Save.

  5. Switch to the Designer tab.

    1. Select the model you want for this tool.

    2. Enter instructions and define parameters.

  6. When your configuration is complete, select Publish. Alternatively, select Save draft if you're not ready to publish this tool.

Configure a standalone computer use tool

The interface for configuring a standalone computer use tool is organized into three tabs to help you manage the lifecycle and observability of your automation.

Overview

The Overview tab acts as the identity card for the tool. Use this tab to:

  • Manage the tool's metadata, such as its name and description.
  • View the current publication status.
  • Monitor recent runs to quickly see how the tool performed in its most recent executions.

Designer

The Designer tab is where you build the logic of your tool.

  • Model: Select the model that powers the automation.
  • Instructions: Write the natural language guidance the AI follows to perform UI tasks.
  • Test: Use the built-in sandbox to run your instructions and see how the AI reacts in real-time.
  • Parameters: Define input and output parameters to pass data into the tool or extract results for downstream steps in a flow.
  • Stored credentials: Securely configure authentication for the websites or apps the tool needs to access.
  • Allowed websites and desktop apps: Define an allow list of specific sites and applications the tool is permitted to interact with.
  • Human supervision: Trigger human-in-the-loop notifications for steps that require manual validation or more information.

Activity

The Activity tab provides deep observability for debugging and optimization. It lists all runs from the last 28 days. For each run, you can review:

  • Session replay: Watch a recording of the AI's interaction with the interface.
  • Screenshots: Review every screen captured during execution.
  • Step logs: See the exact sequence of actions taken by the tool.
  • Metrics: Analyze execution time and success rates.

Integration behavior

You can use a computer use tool in different parts of the platform. The behavior of the tool depends on how you add it.

Integration method Behavior Lifecycle
Add to an agent (as a local tool) The tool is duplicated to the agent's local tools Independent: Changes to the original standalone tool don't affect the agent's version.
Add to an agent flow (as a node) The tool is referenced as a modular node. Linked: The flow always uses the latest published version of the standalone tool.

Add a standalone computer use tool to an agent flow

Note

Agent flows can only access published computer use standalone tools. Agent flows always trigger the latest published version.

  1. In Copilot Studio, select Flows on the side bar.

  2. Select the agent flow you want. The Overview tab for the selected agent flow appears.

  3. Switch to the Designer tab.

  4. Select the Insert a new action icon at the point in the agent flow where you want to invoke your standalone computer use tool. The Add an action panel appears.

  5. Under AI capabilities, select Run a computer use tool. The Run a computer use tool action node appears in the flow.

  6. If you don't have a computer use connection yet, the Create a new connection card appears:

    1. For Connect, select the connection type you want: Machine or Hosted browser.
    2. Provide the required information and select Create new.

  7. Select the computer use tool you want.

  8. If the selected computer use tool has inputs, specify static or dynamic values.

Tip

In the Run a computer use tool action node, you can:

  • Create a new computer use tool by selecting Create in the Computer use tool list.
  • Edit the selected computer use tool by selecting Edit.
  • Refresh your computer use tool list by selecting the Refresh icon.

Add a standalone computer use tool to an agent

Note

When you add a standalone computer use tool to an agent, you create an independent copy embedded within that agent. Updates to the standalone tool don't affect the agent's copy, and vice versa.

  • From the Tools page:

    1. In Copilot Studio, select Tools on the side bar. The Tools page appears.

    2. Select the three dots (…) next to the computer use tool you want, and then select Add to agent. The Add to an agent panel appears.

    3. Select the Agent you want and select Add. The Add tool panel appears.

    4. Select Add and configure. The configuration page for the local copy of your tool appears.

    5. If needed, modify the configuration and then select Save.

    6. Return to the Tools page for your agent and notice the list includes the computer use tool you just added.

  • From the Agent page:

    1. In Copilot Studio, select Agents on the side bar. The Agents page appears.

    2. Select the agent you want.

    3. Go to the Tools page for your agent.

    4. Select Add a tool. The Add tool panel appears.

    5. Select the Computer use tool filter.

    6. Select the computer use tool you want, and then select Add and configure. The configuration page for the local copy of your tool appears.

    7. If needed, modify the configuration and then select Save.

    8. Return to the Tools page for your agent and notice the list includes the computer use tool you just added.

Frequently asked questions

When should I use a standalone tool vs. a local agent tool?

Use the standalone computer use tool when you want to build a single source of truth for a specific agent flow automation that multiple flows need to use. For example, Submit Expense Report. Use a local in-agent computer use tool (duplicated) when the automation requires nondeterministic orchestration governed by the agent itself.

How does the tool handle security and access?

The standalone tool uses two layers of protection:

  • Credentials: Encrypted storage for sign-in data.
  • Governance: A hard restriction (allow list) that prevents the AI from navigating to unauthorized URLs or applications, regardless of the instructions provided.

Can I see what the AI did during a flow execution?

Yes. The Activity tab for the standalone tool lists the session recordings for the last 28 days. These recordings show you every click and keystroke the AI performed, ensuring full transparency and easier debugging.

Why is a model not available in my region?

Anthropic models are currently not available in China. Chinese users should use the OpenAI models.

Related content

  • Last updated on