Share via

Plan and design integration strategies

Agents built in Copilot Studio gain new capabilities through integration with other online services. These services can include Microsoft services, such as SharePoint and Dataverse, third-party services like ServiceNow, Salesforce, and Databricks, and custom APIs. Power Platform offers a rich ecosystem of built-in connectors that are available to Copilot Studio, so you don't need to build them yourself. You also have various options to create custom integrations using low-code or pro-code. Agents can extend their capabilities through the Model Context Protocol (MCP) to interact with external tools and data in a standardized, scalable way, or through computer use to automate web and desktop applications directly from a virtual machine–enabling broader integration scenarios even when no API is available.

By using integrations, agents become a convenient interface for accessing the functionality of multiple systems. Users can perform many different operations without navigating away from the agent. For example, a single agent designed in Copilot Studio can be used to:

  • Read, create, and update items in a SharePoint list.
  • Send an email message by using Office 365 Outlook.
  • Open a new support ticket in ServiceNow.

Integration pattern considerations

Copilot Studio supports integration through Power Platform connectors, HTTP requests, agent flows, bot framework skills, and more. Choosing the correct integration pattern depends on the project's requirements.

Power Platform connectors

Connectors enable agents to interact with an underlying service. They're a set of actions and triggers exposed through a low-code interface. Actions are operations directed by an agent, such as read, write, or update. Triggers notify the agent when an event occurs in another service.

Use prebuilt connectors and custom connectors:

  • Prebuilt connectors: Microsoft offers a large ecosystem of software as a service (SaaS) connectors ready-to-use with Copilot Studio.
  • Custom connectors: When there's no prebuilt connector available, you can build your own connector for a service. They're a no-code or low-code wrapper for REST APIs.

Power Platform connectors meet these requirements:

  • Makers who build with Power Platform are familiar with no-code or low-code connectors.
  • Power Fx code is supported, including variables, conditions, parameters, and more.
  • Parsing and error handling are built in.
  • Prebuilt connectors require only configuration, with no development time needed.
  • Custom connectors can be created once and used across many agents.
  • Ability to monitor activity as part of Copilot Studio integration with Application Insights (a feature of Azure Monitor).
  • Some connectors and custom connectors support virtual networks.

When you use Power Platform connectors for integration, consider the following points:

  • Returning hundreds of results causes a significant delay in the agent's response to the user.
  • Third-party connectors prompt users to enter credentials before they can connect to the service. A user might not understand how to make the connection without guidance. Consider using maker credentials when the user doesn't have the needed information to make the connection.

HTTP requests

Agents can integrate with another service by making an HTTP request. The developer must specify these values each time when making an HTTP request:

  • URI
  • Method (GET, POST, PATCH, and so on)
  • Headers
  • Body

HTTP requests meet these requirements:

  • Ability to connect with services and perform actions not found in the prebuilt connectors list.
  • Less time to develop than a custom connector.

When you use HTTP requests, consider the following factors:

  • Many low-code and no-code makers don't understand how to configure an HTTP request and might have difficulty supporting it.
  • You can't share it across your organization for reuse like custom connectors.

Agent flows

Agent flows execute a series of actions in a predefined sequence. They use the low-code actions found in Power Platform connectors. Agents can pass values as input to an agent flow and receive their outputs.

Agent flows are functionally similar to Power Automate cloud flows with a few important differences:

  • Agent flows can enable high throughput and low latency.
  • Agent flows offer more capabilities than Power Automate flows, such as "human-in-the-loop" actions.
  • Agent flows consume Copilot credits based on their usage. Power Automate flows typically run under a per user licensing model or a per process model.

Agent flows meet these requirements:

  • No-code or low-code.
  • Deterministic chaining of actions.
  • Activity can be monitored separately.
  • Hides sensitive values such as keys and secrets in flow actions. Gets credentials from Azure Key Vault for authentication.
  • Handles large volumes of data and file sizes.
  • Uses concurrency and parallel execution threads to enable better performance.
  • Existing Power Automate cloud flows can be converted to agent flows.

When designing an agent that uses agent flows, consider the following points:

  • Performance is constrained by the API limits of the services and actions it includes.
  • The developer must design an error-handling pattern so the agent knows how to handle exceptions.
  • The response to the agent by the flow has a size limit.
  • Longer-running logic can continue to execute after the Respond To Agent action.

Bot Framework skills

Skills are reusable conversational skill building blocks that you build by using the Bot Framework. Implement skills by using pro-code. Register skills in Copilot Studio to use them. After you register a skill in Copilot, you can trigger it as a node in a conversation topic.

Bot Framework skills meet these requirements:

  • Synchronous execution.
  • Support for private endpoints.
  • Existing Bot Framework skills can be reused in Copilot Studio.

When you use Bot Framework skills, consider the following factors:

  • Requires pro-code for ongoing development and support (for example, C#).
  • Runs in the Azure AI Bot Service.
  • You incur added costs in Azure that you must cover with an Azure subscription.
  • Application Lifecycle Management (ALM) is outside Power Platform.

Other considerations for integration patterns

Along with the guidance for each integration type, consider the following points for an overall integration strategy:

  • Copilot Studio integration patterns aren't exclusive and can be combined.
  • Integrations are only as fast as the endpoints they connect to.
  • Always optimize queries to return results as fast as possible to provide a good conversational experience.
  • Where agent flows and Copilot Studio don't execute logic fast enough, consider these options:
    • Dataverse custom APIs
    • Dataverse low-code plugins
    • Azure functions
  • Specific scenarios require the data integration to be performed in a middle-layer that modifies messages as it relays them.

Adding integrations to an agent

Add integrations to an agent by using the following methods:

  • Tools

    • Connect the integration as a tool to expand an agent's functionality.
    • Use natural language to describe when the orchestrator should select the tool to fulfill a task.
    • Define how the agent should determine the tool's input values and what the output should look like. Generative AI can dynamically fill in inputs based on a description, or you can manually define values and formulas.
    • Invoke as part of a plan created by generative orchestration.

  • Topics

    • Call an integration through a topic node. Executes at the same position in the topic's sequence each time.
    • Provide inputs to the node by using Power Fx.

  • Agents

    • Create a child agent that has its own set of instructions and define access to knowledge and tools available to the parent agent.
    • Alternatively, connect to an existing agent in Copilot Studio, Microsoft 365 SDK, Microsoft Foundry, Fabric, Agent-To-Agent, and so on.

Quotas and limits

Review current service limits for Copilot Studio and Power Platform:

You must determine API limits for these services separately:

  • OpenAI capacity: Rate limits aren't documented.
  • Bot Framework skills: Check the Azure documentation for the specific services used.

Integrations planning overview example

When planning integrations with Copilot Studio, determine the following information for each service:

  • Connected system: The service you're connecting to.
  • Called by: The bot component in Copilot Studio that calls the service.
  • Called with: The selected integration type.
  • Expected daily volume: The average number of requests.
  • Expected peak: The highest anticipated number of daily requests.
  • Details: Other information about the integration.

Here's an example for an agent with multiple integrations:

Connected system Called by Called with Expected daily volume Expected peak API calls per minute Details
ServiceNow Service Desk KB topic Workflow 5,000 1,000 600 The user query is used as an input to run a query on ServiceNow knowledge base by using the standard connector. A JSON of results is returned to Copilot Studio for generative actions custom data.
Contoso website Knowledge Generative answers 1,000 100 N/A https://contoso.com/en-us/FAQ with /en-us/ being dynamically set based on the user locale variable.
Internal Directory API Conversation Start topic Workflow 15,000 5,000 N/A Use of custom connectors running in the context of an Azure Virtual network to connect to an internal API.
Weather API Weather topic HTTP 500 100 N/A GET request to a REST-based API.
Teams Meeting Booking topic Tools 1,500 300 100 Create a Teams meeting based on conversation inputs.
SAP Every 24 hours Workflow 20,000 - 2,500 Daily batch synchronization of the product catalog from SAP to Dataverse.

Next step

Learn how AI prompts, Model Context Protocol, and computer use agents can extend your agent's capabilities.

Use agent tools to extend, automate, and enhance your agents

  • Last updated on