Copilot Studio adds MCP tools for agents

Key insights

• The video explains how to add MCP tools to Microsoft Copilot Studio agents.
• MCP stands for Model Context Protocol and exposes tools from external services.
• The setup starts in the agent’s Build tab under Tools and Add a tool.
• Copilot Studio reviews the discovered tools before you save the configuration.
• The approach is meant to extend agents beyond basic chat and into third-party systems.

Griffin Lickfeldt (Citizen Developer) recently published a short tutorial on how to add MCP tools to agents in Microsoft Copilot Studio. The video’s main point is straightforward: MCP gives agents a way to use external tools so they can do more than answer questions.

The description frames the topic as a “Copilot Studio for Noobs” walkthrough, so the focus is on the basic setup rather than advanced design patterns. Even so, the subject matters because MCP connects agents to outside systems in a standardized way, which can change how makers think about agent capabilities.

What's new

The video is about using Model Context Protocol, or MCP, as a tool source for Copilot Studio agents. In the material provided, MCP is presented as a way to extend an agent with external tools, connect to third-party systems, and add advanced capabilities.

The research notes add that Copilot Studio now has a native MCP server experience in the agent designer. That means makers can start from the Build tab, open Tools, and add an MCP source without treating every integration as a one-off custom connector.

Another detail from the notes is that Copilot Studio can review the tools exposed by the MCP server after the connection is made. That is important because the agent does not just connect blindly; it can see what the server offers before the configuration is saved.

Why it matters

For makers, MCP is relevant because it offers a more structured way to connect an agent to outside capabilities. Instead of building separate actions for every task, they can connect to a server that publishes multiple tools.

For admins and solution owners, the value is in standardization. The notes say MCP can work with external services, databases, APIs, and business systems, so the same pattern can cover several integration scenarios. That can make agent design easier to reason about in a low-code environment.

For end users, the practical effect is that agents can do more than basic chat. The video says the goal is to build agents that go beyond chat, which suggests more useful conversational workflows when an agent has access to the right tools.

If you track Developer Tools content, this sits in the same broader space as low-code integration work. It is not about a full app rewrite; it is about giving an agent controlled access to outside functions.

How it works

The tutorial material describes the setup at a high level, and the research notes fill in the steps. The process begins in Copilot Studio with the relevant agent open.

1. Open the agent in Copilot Studio.
2. Go to the Build tab.
3. Open Tools in the components panel.
4. Select Add a tool.
5. Choose Model Context Protocol (MCP).
6. Enter the server details, including the server URL and authentication information.
7. Review the tools that Copilot Studio discovers from the server.
8. Save the configuration.

The notes say the configuration can include a name and description for the server, plus HTTPS endpoint details and the authentication method. They also say Copilot Studio performs a protocol handshake with the server and then shows the available tools for review.

That matters because the agent is not just linked to a server name. It is connected to a server that declares what it can do, and Copilot Studio exposes those capabilities as tools the maker can evaluate.

Limitations and open questions

The source material is short and does not show the full user interface, error handling, or permission model. It also does not explain how long setup takes, what specific authentication methods are supported, or whether every organization will see the same MCP experience.

The notes mention that MCP can appear through Microsoft-provided MCP connectors or through existing custom MCP services, but they do not list the exact service types or deployment requirements. They also say the onboarding wizard is the recommended method, while custom connectors remain an alternative, but the material does not compare those paths in detail.

Because this is a short video description and summary, the exact limits of tools versus resources are also only touched on. The notes say Copilot Studio distinguishes between the two, but the source material does not explain every use case or edge case.

In other words, the material confirms the direction of travel, not every operational detail. If you need a production rollout decision, you would still want to verify current Microsoft documentation and your tenant’s support status.

Sources

• Griffin Lickfeldt video: How to Add MCP Tools to Agents in Copilot Studio
• HubSite 365 publisher website

Frequently asked questions

What is MCP in Copilot Studio?

The material describes MCP, or Model Context Protocol, as a standard for exposing tools and resources to AI agents. In this context, it lets a Copilot Studio agent use capabilities from an external server.

Where do you add MCP tools in Copilot Studio?

The notes say you start in the agent’s Build tab, open Tools, and choose Add a tool. From there, you select Model Context Protocol (MCP) and enter the server details.

What can an MCP server provide?

The research notes say an MCP server can expose tools for querying business data, calling APIs, searching knowledge sources, starting processes, and updating connected systems. The source material does not list specific servers or products.

Does the material explain authentication and rollout limits?

Only partly. It says authentication details are entered during setup, but it does not specify which methods are supported or whether the feature is available in every tenant.