Introduction
To support business work, an agent needs access to the information people rely on and the systems where they take action. When you design an agent powered by the GitHub Copilot harness in Microsoft Copilot Studio, part of the work is recommending how the agent connects to that information and those systems. The right recommendation depends on what the business needs the agent to do: answer from content, read or change data in another system, or hand work to a specialist.
Suppose you work on the operations team at a regional electric utility. Staff who handle service cases lose time switching between policy documents, the case system, and specialist reviewers. Your team asks you to recommend how one agent should connect to these resources so staff can get current policy explained, look up a case, submit an approved change, and refer unusual exceptions to a specialist agent.
In this module, you explore:
- How knowledge, tools, and connected agents each meet a different kind of need.
- How to compare agent tools: a connector, a Model Context Protocol (MCP) server, a workflow, or Foundry IQ, Fabric IQ, or Work IQ.
- The identity and access checks that determine whether an approach is viable.
- How to help the agent choose the right capability.
By the end of this module, you're able to recommend a supported integration approach for an agent powered by the GitHub Copilot harness, explain its access constraints, and draft descriptions that help the agent choose the right capability.