| tags | MS-4022, Reference |
|---|---|
| GA | G-DXYJBX6BH8 |
Build declarative agents with Microsoft Copilot Studio that extend Microsoft 365 Copilot with your own knowledge, prompt tools, and connector tools, then publish them to the people in your organization.
:::success Date: 20260820 Course ID: 103821 :::
:::info Course Survey: https://aka.ms/ms4022survey :::
Course MS-4022 English version
:::success Training key: DE5F5F7CA5034DFF :::
Declarative agents for Microsoft 365 Copilot
Write effective instructions for declarative agents
Add SharePoint as a knowledge source
Add knowledge to an existing agent
Extend Microsoft 365 Copilot with the Copilot chat harness
Use shared tools from the Tools page
Create and manage custom solutions
Pass files to agent flows, connectors, and tools
Extend your agent with Model Context Protocol
Distribute your prompt using a solution
Use connectors in Copilot Studio agents
SharePoint connector reference
Power Hour: Reimagine Automation with Copilot Studio Workflows
:::info These two links sit outside the five course modules. Copilot Studio workflows are an automation capability for custom agents rather than for the declarative agents built in this course, and the blog is the quickest way to track product changes after class. :::
| No. | Name | Link |
|---|---|---|
| 01-01 | Introduction to agents | Watch |
| 01-02 | What are Microsoft 365 Copilot agents and how to use them | Watch |
| No. | Name | Link |
|---|---|---|
| 02-01 | Microsoft 365 Copilot | Copilot Studio agent builder | Watch |
| 02-02 | Deploy a declarative agent for Microsoft 365 Copilot | Watch |
| No. | Name | Link |
|---|---|---|
| 03-01 | Integrate with MCP servers | Mission 10 | Agent Operative | Watch |
| No. | Name | Link |
|---|---|---|
| 04-01 | Extracting resume contents with multi-modal prompts | Mission 7 | Agent Operative | Watch |
# Extend Microsoft 365 Copilot in Copilot Studio
## M01 - Introduction to declarative agents for Microsoft 365 Copilot
### Declarative agents
- Chat-based agents that provide information and complete tasks, a good fit for customer support, IT service desk, and HR support
- Use one immersively in a 1:1 chat, or @mention it to keep the context of the current Copilot conversation
### Three building blocks
- [Instructions](https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/declarative-agent-instructions) define behavior, tone, boundaries, and fallback
- Knowledge grounds the answers; tools let the agent reach external systems
### When to choose one
- The scenario is well defined and the Microsoft 365 Copilot model and orchestration are enough
- The data already lives in Microsoft 365, or a connector or tool can reach it
## M02 - Build your first declarative agent for Microsoft 365 Copilot by using Copilot Studio
### Create and test
- Build the Product Support agent from the Microsoft 365 Copilot agent area in Copilot Studio
- Write instructions that cover capabilities, tone, information sources, and fallback
### Custom knowledge
- A [SharePoint knowledge source](https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-add-sharepoint) grounds answers and respects each user's permissions
- Answers should carry citations; when nothing is found, follow the fallback defined in the instructions
### Publish and interact
- Suggested prompts show common tasks and the range of what the agent can do
- After publishing to Microsoft 365 Copilot, validate both the immersive and the in-context experience
## M03 - Introduction to tools for declarative agents in Copilot Studio
### What a tool is
- A tool is a reusable capability; one core tool can be wrapped per agent with different inputs, outputs, and configuration
- Ask first: does the agent need to read, update, run a process, or generate text?
### Tool types
- Prompt tools run a predefined prompt; connector tools read or update external data through an API
- REST API and [MCP](https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-extend-action-mcp) widen the connection options; agent flows are not supported for declarative agents
### Configuration
- Clear descriptions help orchestration match user intent and inputs to the right tool
- Plan the Power Platform environment, solution, authentication, and input and output variables first
## M04 - Extend declarative agents with prompt tools in Copilot Studio
### Prompt tools
- A reusable custom prompt template for classifying information, extracting entities, drafting replies, and summarizing content
- Inputs carry data at run time; outputs define what the tool returns
### Prompt engineering
- Be specific, show examples, keep it simple, and say what to do when the task cannot be completed
- Test with sample data in the prompt builder, review the model response, and refine
### Publish and use
- Add the finished prompt tool to the agent, then test that a user question produces the expected format
## M05 - Extend declarative agents with connector tools in Copilot Studio
### Connector tools
- A Power Platform connector wraps an API so the agent can reach Microsoft and external services
- Standard, premium, and custom connector availability depends on the plan and on organization settings
### Describe the action clearly
- The tool name must be unique and recognizable; the description explains the action and the expected input and output
- A clear description is what lets the orchestrator match a user message to the right action
### Product Support example
- Use the SharePoint Get items connector tool to query product support cases in the Support Cases list
- Users connect with their own credentials, so the agent never exceeds their data permissions
- Money Yu
- Mail: Money.Yu@microsoft.com
- LinkedIn: @abc12207