Model Context Protocol (MCP) is now generally available in Microsoft Copilot Studio

MCP now includes a new set of features and enhancements that support more robust and scalable deployments: tool listing, enhanced tracing, and more.

When we introduced Model Context Protocol (MCP) integration in public preview, our goal was simple. We wanted to provide a standard, reliable way to allow you to bring your external data tools and knowledge into Microsoft Copilot Studio, offering easier integration and greater flexibility.

Today, we’re thrilled to announce the general availability of MCP integration in Copilot Studio! With MCP, you can add AI apps and agents into Copilot Studio with just a few clicks, empowering makers to seamlessly integrate Copilot Studio with existing knowledge sources and APIs.

By connecting to an MCP server, agents are instantly equipped with the latest actions and information – automatically updated as systems evolve. The result is a faster, smarter way to build and scale agents, dramatically cutting down on manual upkeep and accelerating innovation.

New enhancements and capabilities

Now that MCP is generally available, it includes a new set of features and enhancements that support more robust and scalable deployments:

Tool listing: The MCP server settings page now provides users a clear and organized view of all available tools included with the MCP server. This improvement increases transparency, making it easier to explore and manage the full range of tools associated with your connection.
Streamable transport: We’ve expanded our transport layer to support streamable data transfer. The focus is on optimizing and staying up to date with the latest protocol version that aligns better with your deployment needs. Note that since SSE transport has been deprecated, SSE transport support in MCS will remain in public preview.
Enhanced tracing and analytics: To provide deeper insights and improve debugging capabilities, we’ve significantly enhanced our tracing and analytics tools. The activity map in Copilot Studio will now allow you to see which MCP server and specific tool within was invoked at runtime.
Quality improvements: As part of our ongoing commitment to enhancing the user experience, we’ve implemented several quality improvements across the platform. These improvements include performance optimizations, bug fixes, and reliability enhancements. All these help your deployments run more smoothly and efficiently, with fewer interruptions.

Why use MCP?

MCP simplifies the integration with AI apps and agents, empowering makers to:

Seamlessly integrate with data sources: Whether you’re using internal APIs or third-party services, MCP ensures dependable and straightforward integration within Copilot Studio.
Leverage a library of remote MCP servers: Beyond building custom integrations, users can access a growing marketplace of certified MCP servers. This makes it quicker and easier to connect with various tools.
Enable dynamic and adaptable functionality: MCP servers can supply tools and data to agents in real time. This offers more flexibility while minimizing the effort and cost of integration and maintenance.

For example, integrating a banking MCP into an agent instantly unlocks a range of additional capabilities – without the overhead of manually configuring each action.

To get started, access your agent in Copilot Studio, select ‘Add a Tool,’ and search for your MCP server! (Note: generative orchestration must be enabled to use MCP.) Integrating your own Model Context Protocol (MCP) server with Copilot Studio involves three main steps:

Build the server: Start by creating a server using one of the available SDKs. This server acts as the core for managing your tools and data. You can customize it to meet your specific requirements – such as supporting custom data formats, model types, or unique workflows.
Create a connector: After the server is ready, the next step is to build a custom connector that links your MCP server to Copilot Studio.
Connect and use in Copilot Studio: Once your connector is in place, you can begin accessing your data and interacting with your models directly within Copilot Studio.

We hope the new improvements enhance your experience as you explore MCP. We’re looking forward to your valuable feedback as you continue to use and grow with the platform.