Microsoft Agent 365 SDK Overview

The Microsoft Agent 365 SDK extends your existing AI agent built on any agent stack with enterprise-grade identity, observability, tools, and governance.

With the Agent 365 SDK, developers can optionally utilize Agent Identity to receive and respond to notifications from productivity apps. The SDK also enables agents to invoke governed and observable Model Context Protocol (MCP) servers from Microsoft. Additionally, agents can emit telemetry to provide IT admins with insights into agent functionality and performance.

What is the Agent 365 SDK?

The Agent 365 SDK is a comprehensive software development kit that extends agents to fit seamlessly into existing workflows, while still functioning independently when required. Agents can have unique identities, can be invoked using common gestures (such as @mentions) in apps that users typically operate in (such as Teams, Word, Outlook, etc.), demonstrate observable behaviors, building trust, taking auditable actions, and achieving it all via secure access to tools and data.

Agent Identity allows for gated, monitored access control to an administrator while still providing the agent with the power it needs to thrive. Agents acquire an identity through the acquisition process. During the development lifecycle, developers can obtain a Microsoft Entra App Blueprint with the required MCP servers using the Agent 365 CLI and create the template for all agent instances created from that blueprint.

With notifications, we bring agents' interactions patterns closer to a human. Agents can respond to users in context within productivity apps such as Teams or to comments in Word. Agents can also respond to emails sent to the agent's mailbox, which are delivered to a notification endpoint hosted by the agent. The agent can also participate in lifecycle operations such as Acquire and UnAcquire and choose to do set up and cleanup cycles during that time for an Agent Identity.

Built on and integrated with Open Telemetry, the Agents 365 SDK captures detailed telemetry for each agent's activity, including inference events, tool usage, and notifications. Administrators gain full visibility into how agents operate and interact with users, systems, and data using Purview and Defender.

Agents can utilize the Microsoft MCP servers to carry out various scenarios such as sending email, locate documents in SharePoint, and respond to the comments in a document with minimal changes to the Agent code. Administrators govern these MCP servers and the growing list of the tools empower the agent to thrive in the productivity space.

Agent 365 works with agents built on Microsoft Agent Framework, Copilot Studio, Azure AI Foundry, Semantic Kernel, OpenAI Agents SDK, Claude Code SDK, and LangGraph.

How is Agent 365 SDK different?

• Agent 365 SDK isn't an agent stack. It isn't the way to create or host an agent and doesn't come with any orchestrator or workflow management contrary to Microsoft Agent Framework or Microsoft Copilot Studio or Microsoft Foundry.
• It doesn't replace Microsoft 365 Agent SDK and the functionality it provides for hosting, activity protocol, and identity. On the contrary, Agent 365 SDK uses packages for Microsoft 365 Agent SDK to deliver notifications via the Activity protocol and uses the Agent Identity packages to obtain the Agentic token for an agent when applicable.
• Agent 365 asks developers to start with an existing agent when available from an agent framework of their choice. They can then extend their agent to provide more functionality that makes their agents governable, observable, and more productive in the digital world.

Understanding the blueprint of an agent in Agent 365

Before creating an instance of your agent, it's helpful to understand what we mean by the blueprint of an Agent 365 agent. A blueprint is an IT-approved, preconfigured definition of an agent type, essentially the enterprise "template" from which all compliant agents are created. It defines the agent's capabilities, required MCP tool access, security and compliance constraints, audit requirements, lifecycle metadata, and any linked governance policy templates such as DLP, external access restrictions, or logging rules. When a blueprint is activated, it becomes hirable, meaning administrators can instantiate new agents from it in the Microsoft 365 Admin Center. Every agent instance inherits its blueprint's rules, ensuring consistent governance across mail, calendar, Teams messaging, SharePoint access, and any other Microsoft 365 workloads the agent touches. This prevents "shadow agents," enforces observability, and anchors each agent in a secure Entra-backed identity and permissions model.

Agent 365 Agent SDK packages

Find the Agent 365 Agent SDK packages in the package managers for the supported languages.

High-level flow of an Agent 365 Agent

The following diagram illustrates the developer flow for agents

You start with an agent you've already built using a programming language and the agent framework of your choice. You can then use the Agent 365 SDK to extend the agents you've already created, adding the specialized capabilities such as notifications, observability, and tool server connections they need to run within the platform. If you want examples of how existing agents are structured, the quickstarts walk you through the sample agents included in the Microsoft 365 Agents Toolkit for VS Code. See the available Agent 365 quickstarts for examples:

• Quickstart: Setup and run the JavaScript LangChain sample agent
• Quickstart: Setup and run the Python Agent framework sample agent
• Quickstart: Setup and run the JavaScript Claude Agent SDK sample agent

You determine which Microsoft MCP servers your agent relies on, configure those connections, and integrate the Agent 365 components - observability, notifications, and tool server connections, using the patterns provided in the GitHub samples.

After that, generate an Agent 365 blueprint that defines how your agent is represented and wired into the platform. With the blueprint in place, test your agent in the Agents Playground. This gives you a space to validate notifications, try out MCP interactions, and inspect telemetry before deploying anywhere.

When everything looks good, publish your agent to a tenant. From there, administrators can enable and manage it in the Microsoft 365 admin center.

When do you need the Agent 365 SDK?

Use the Agent 365 SDK when you need to:

• Integrate with Microsoft 365 data or tools via secure MCP connectors
• Govern and audit multiple agents across your organization
• Provide enterprise-grade telemetry and monitoring
• Enable notifications for agent interactions
• Deploy in regulated environments requiring traceability and lifecycle management

How the Agent 365 SDK fits into the agent ecosystem

Benefits at a glance

• Unified identity and governance across all agents
• Out-of-the-box telemetry and tracing
• Secure and governed access to Microsoft 365 tools
• Cross-orchestrator and multicloud support
• Simplified upgrade path for existing agents

Next steps

• Learn about agent blueprint and instance setup
• Add observability to your agent
• Add and manage tools