The Model Context Protocol (MCP) is revolutionizing AI by providing a universal integration standard for the external systems, unlocking unprecedented capabilities, transforming how AI models access and utilize real-time data. In keeping with this trend, at Ignite 2025 we announced the public preview of Model Context Protocol (MCP) support in declarative agents.  

Declarative agents enable you to customize Microsoft 365 Copilot to help you meet the unique business needs of your users. With introduction of MCP support, it’s now much easier for developers to integrate their business workflows, SaaS, and LoB systems into Copilot via declarative agent.  

Let’s dive into how you can create your declarative agent using MCP server.  

Build your own MCP-based agent with Microsoft 365 Agents Toolkit

You can start building your own MCP-based declarative agent using the Microsoft 365 Agents Toolkit in Visual Studio Code. The toolkit provides a guided, simple yet powerful point-and-click experience to connect your agent to an MCP endpoint.  

To build an MCP-based agent: 

1. Scaffold a new agent: In Visual Studio Code, choose Microsoft 365 Agents Toolkit, and select Create a new Declarative Agent. This sets up the project structure and manifest files for your Copilot agent.  
2. Add the MCP action: Choose Add Action –> Start with an MCP server in the toolkit. You’re prompted to enter your MCP server URL (usually https://<your-server>/mcp). That’s all you need – the toolkit will fetch the server’s list and description of its tools and automatically generate a plugin spec from it.  
3. Choose your tools: If your MCP server exposes many functions, you can select a subset to include in your agent (for security or simplicity). The toolkit lets you pick which tools to import and tools names, parameters, and definitions are defined in the manifest.   
4. Configure authentication: If the MCP server requires auth (which most do), the toolkit walks you through adding credentials. Currently SSO and static OAuth 2.0 is supported by default.  
5. Generate & review: After you provide the configuration information, the Agents Toolkit generates all the files and scaffolding. It updates the app manifest manifest.json to list the plugin, fills in the ai-plugin.json with the MCP actions, and adds the action into declarativeAgent.json . Essentially, it writes the “wiring” for you – no manual JSON editing required. You can review these files to see the imported actions and make sure everything looks right. 
6. Test and publish: With everything set, you can run the agent locally (the toolkit has a one-click Provision and Start debugging feature that sideloads your agent for testing in the Microsoft 365 Copilot app). Because the toolkit took care of registration and auth, your agent can immediately call the MCP server’s tools in response to user prompts. 

Optionally, you can define response semantics in the form of Adaptive Cards and write agent descriptions to improve relevance as well as tool chaining. To learn more about response semantics, see Adaptive Card response templates.

These steps require little code writing or manual editing to get started. The heavy lifting – reading the MCP schema, creating function definitions, hooking up OAuth – is automated by the toolkit, with you in full control.   

Publish MCP-based declarative agents

When you’re ready, choose how you want to distribute your agent. You can either: 

• Distribute via Agent store: Submit the package in Partner Center and it will go through the validation process. When it’s approved, it’s published in the Copilot store. For more information, see Publish agents.
• Deploy in your organization: Your IT admin can deploy these agents to all employees or select employees based on use case. For more information, see Manage Copilot agents. 

IT admins can manage MCP-based agents just like REST API agents.  

Partners pioneering MCP-based agents

A number of partners have already embarked on a journey to build MCP-based declarative agents. These agents provide a glimpse of what’s possible.  

monday.com 

monday.com is a flexible WorkOS that teams use to plan, run, and track their work in one connected place. With the monday.com agent inside Copilot, teams deliver results across projects, marketing, sales, operations, and engineering, all through natural language, without switching tools. 

The agent enables: 

• Project setup that starts instantly: Use monday’s MCP-powered agent to create new boards the moment a project is named, complete with custom columns like status, owners, dates, and more. 

• Faster task workflows: Add items, assign work, update statuses, or adjust timelines directly through Copilot commands, keeping teams aligned and unblocking progress in real time. 

• Clear insights and more intelligent decisions: Pull summaries, activity logs, and item-level detail to understand progress and spot risks. For example: “Show me all the tasks in ‘Blocked’ and who owns them.” 

Get started with the monday.com agent in Copilot and bring AI-powered execution directly into your workflow.

Canva

Canva, the world’s all-in-one visual communication platform, is bringing its design tools into the flow of conversation in Copilot. This makes it seamless to create design content in context, via the Copilot agent. 

• Content generation: Generate presentations, social posts and more with Canva AI.
• Asset management: Browse, search. and summarize existing Canva designs and assets in Copilot.

Try the Canva agent in Copilot today.

Sitecore 

Sitecore empowers global marketing teams to launch high-impact campaigns and experiences, personalize content for any audience, and optimize every digital touchpoint – all powered by the Microsoft ecosystem. It recently integrated its Marketer MCP with Copilot to empower marketers with fast, intelligent access to take action within Microsoft 365. This means: 

• Unified workflows in one interface: Eliminate tool-switching by creating pages, localizing campaigns, personalizing experiences, and running tests all in one place. 

• Agentic marketing: Work directly in Copilot to drive campaigns and utilize agents to multiply marketing output. 

• Enterprise-ready foundation: Confidently scale marketing with security, reliability, and global reach. 

Looking ahead 

The introduction of MCP support, along with REST APIs and message extensions in declarative agents, is a significant milestone for Copilot extensibility. It combines the standardization and power of MCP with the user-friendly, declarative model of Copilot agents, giving developers the best of both worlds. We’re already seeing increased velocity in agent development and innovative scenarios from our preview partners, and we expect many more to emerge as MCP enters broader use. 

If you have an MCP server (or plan to create one), now is the time to build an agent and integrate it with Copilot. The barrier to entry has never been lower – you can easily plug your service into Copilot with just a URL and a few configuration settings. The Agents Toolkit walks you through it, and you can also find step-by-step guidance. Our goal is to make MCP integrations seamless and robust for both developers and users. 

Happy agent building, and as always, let us know your thoughts through feedback in Copilot or leave a comment on this post as you explore MCP-powered declarative agents!