Introduction
An agent becomes more useful when it can use approved information, retrieve current data, and follow reusable task guidance. In Microsoft Copilot Studio, these capabilities come from different building blocks. Choosing the right building block helps the agent complete its task without adding unnecessary complexity.
Grounding means connecting an agent's response to information from an approved source instead of relying only on the model's general knowledge. A tool lets an agent retrieve current data or perform an action. A skill gives the agent reusable instructions for completing a specific kind of task.
In this module, you learn how to:
- Distinguish knowledge, tools, and skills.
- Choose between connector, Model Context Protocol (MCP), and workflow tools.
- Add file, website, SharePoint, or Dataverse knowledge.
- Disable broad web search when the agent should use only approved sources.
- Validate grounding and citations.
- Configure an existing connector tool and its inputs.
- Connect an agent to an existing MCP server.
- Create or add a skill and update its activation guidance or instructions.
- Select an agent model and configure content moderation.
- Test tools and improve an agent after observing incomplete or incorrect behavior.
The module uses agents powered by the Microsoft Copilot harness. It focuses on configuring models, moderation, knowledge sources, tools, and skills in Copilot Studio. Connecting, authenticating, and troubleshooting custom MCP servers requires advanced integration skills and isn't part of the hands-on exercise.
Note
Building and testing agents powered by the Microsoft Copilot harness can consume Copilot Credits before publication. Confirm that your environment has approved access and sufficient credit capacity before you begin. The exercise also requires a SharePoint site where you can create a list and permission to use the SharePoint connector. For more information, see Overview of usage-based billing.