The Model Context Protocol (MCP) is changing the way AI interacts with external systems, and now, Microsoft 365 Copilot developers can harness its power through declarative agents. With MCP, developers get a universal integration standard that allows AI models to access and act on real-time data, streamlining the process of bringing business workflows, SaaS, and line-of-business (LoB) systems into Copilot’s conversational interface.
Getting Started: Building MCP-Based Declarative Agents
Developers can quickly create MCP-based declarative agents using the Microsoft 365 Agents Toolkit in Visual Studio Code. This toolkit offers a guided, point-and-click experience with no extensive coding required. Here’s how the process unfolds:
- Scaffold a new agent: Initiate a new declarative agent project, which automatically sets up the necessary structure and manifest files.
- Add MCP actions: Connect to your MCP server by entering its endpoint. The toolkit fetches available tools and generates a plugin specification.
- Select tools: Choose specific tools or functions exposed by the MCP server to import, which is useful for tailoring security and functionality.
- Configure authentication: Add credentials, with support for SSO and OAuth 2.0 out of the box, ensuring secure connections.
- Generate and review: The toolkit auto-generates all required configuration files, wiring up your agent without manual JSON edits.
- Test and publish: Use built-in debugging to run and test your agent locally in Copilot before publishing to your organization or the Copilot store.
Enhancements like AdaptiveCard response templates are available to improve response semantics and tool chaining, ensuring agents deliver relevant, actionable outputs.
Publishing and Managing Agents
Once your agent is ready, Microsoft offers flexible deployment options:
- Copilot Agent Store: Submit your agent via Partner Center for validation and, once approved, publish it to the Copilot store for broad distribution.
- Organizational Deployment: IT admins can deploy agents organization-wide or to select users, seamlessly managing MCP-based agents alongside REST API agents.
Innovative Use Cases from Early Partners
Several partners have already showcased what’s possible with MCP-powered Copilot agents, delivering real value through seamless natural language integration:
- monday.com: Teams can instantly set up projects, manage workflows, and gain insights, all within Copilot, empowering users to handle tasks, assignments, and updates in real time without leaving their workflow.
- Canva: Copilot users can now generate content, create presentations, manage design assets, and summarize information from Canva by conversing with the agent.
- Sitecore: Marketing teams benefit from unified campaign management, content personalization, and enterprise-grade scalability, accessible through Copilot’s conversational interface.
The Road Ahead for MCP and Copilot Extensibility
The addition of MCP support to declarative agents marks a significant leap for Copilot extensibility. It combines the standardization and power of MCP with Copilot’s easy-to-use, declarative approach, enabling rapid agent development and unlocking new scenarios for both developers and users. The barrier to entry has never been lower: with just a URL and a few configuration steps, developers can plug their services into Copilot. Comprehensive tooling and documentation ensure that integrations are both seamless and robust.
The Agents Toolkit walks you through it, and you can also find step-by-step guidance.
Conclusion
MCP-based declarative agents are transforming how developers and organizations integrate their systems with Microsoft 365 Copilot. By simplifying the build and deployment process, MCP empowers teams to bring powerful, AI-driven capabilities directly into their daily workflows. Now is the ideal time for developers to explore MCP integration and contribute to the growing ecosystem of intelligent Copilot agents.
Source: Microsoft 365 Developer Blog
How MCP Declarative Agents Are Transforming Copilot Integration