Extending Copilot Studio with Plugin Actions

1. From Answering to Acting: Introducing Plugin Actions

Plugin actions are reusable components—like Power Automate flows, custom connectors, or even pre-built Power Platform connectors—that your copilot can intelligently call upon to fulfill a user's request.

The key technology that makes this so powerful is Dynamic Chaining. Unlike classic bots where you had to define rigid trigger phrases for every topic, dynamic chaining allows the copilot's underlying Large Language Model (LLM) to:

1. Infer the user's intent from natural conversation.
2. Identify the correct plugin action to call from the available tools.
3. Extract the necessary parameters from the user's message.
4. Prompt the user for missing information if needed.
5. Execute the action.

This architecture allows for much more fluid and natural conversations, where the copilot can seamlessly transition between providing information and performing actions.

The user interface for adding these actions makes the concept clear. You select from available "tools" to grant your copilot new abilities.

To begin, let's watch a short video that explains the concept of plugins and the magic of dynamic chaining.

2. Extending with Custom Connectors

Given your extensive background in C# and developing with REST APIs, using custom connectors will be a natural way for you to extend Copilot Studio. A custom connector is essentially a wrapper around a REST API that allows it to be used easily within the Power Platform, including as a plugin action for your copilot.

The process involves these main steps:

1. General: Provide basic information like an icon, description, and the API's host and base URL.
2. Security: Define the authentication method (e.g., API Key, OAuth 2.0).
3. Definition: Define the actions (API operations) your connector can perform. For each action, you specify the request (verb, URL, headers, body) and a sample response. This is analogous to defining the method signatures and data contracts in a C# interface.
4. Test: Validate the connector to ensure it works as expected.

For a detailed, step-by-step guide you can refer to later, the Microsoft Learn article "Create a custom connector from scratch" is an excellent resource.

Now, let's see this in action. The following video segment demonstrates building a plugin action based on a custom connector that connects to a custom API for managing orders.

3. Extending with Power Automate (Agent Flows)

While custom connectors are perfect for exposing a specific API, you'll often need to perform more complex, multi-step business logic. This is where Agent Flows come in. An Agent Flow is a special type of Power Automate cloud flow that is designed to be called by a copilot.

You should choose an Agent Flow over a custom connector when you need to:

• Orchestrate a sequence of actions across multiple services (e.g., Dataverse, SharePoint, Outlook, Teams).
• Implement complex conditional logic (if/else, switch cases).
• Leverage the vast library of over 1000 pre-built connectors available in Power Automate.
• Incorporate AI Builder actions or advanced approval processes within your automation.

Agent Flows are created directly within the Copilot Studio interface and are solution-aware by default. A key difference from standard flows is that they are triggered by the agent and can receive inputs from, and return outputs to, the copilot conversation.

The following practical lab guide will walk you through building an Agent Flow that posts an Adaptive Card to a Teams channel. While the full lab describes an autonomous agent, we will focus on the part that is directly relevant to our learning outcome: creating a flow that is used as a tool.

Conclusion

In this lesson, we have significantly expanded our copilot's abilities, moving it from a passive information provider to an active participant in business processes. You've learned how to connect it to the outside world and give it the power to execute tasks.

Key Takeaways:

• Plugin Actions are the mechanism for enabling a copilot to perform actions by calling tools like custom connectors and flows.
• Dynamic Chaining is the AI-powered feature that allows the copilot to intelligently understand user intent and invoke the correct plugin without needing rigid trigger phrases.
• Use Custom Connectors to expose any RESTful API to your copilot, which is ideal for single-purpose interactions with external or custom-built systems. Your existing API development skills are directly transferable here.
• Use Agent Flows (Power Automate) when you need to orchestrate complex, multi-step logic that involves multiple services and connectors within the Power Platform ecosystem.

What's Next?

You have now built a topic with manual triggers, integrated generative answers for dynamic Q&A, and added plugin actions to perform tasks. You have all the core components of a sophisticated copilot. In our next lesson, we will focus on bringing it all together by learning how to Test, debug, and publish a chatbot to a channel like a custom website. We will explore tools like the Monitor to trace conversations and ensure all your powerful new features work together seamlessly before you deploy your copilot for users.