Editor & agent
Install VS Code and GitHub Copilot. Use Agent mode.
- Add Copilot CLI for terminal-based plugin installation.
- VS Code supports
@agentPluginsin Extensions and discovers CLI-installed plugins.
GitHub Copilot × Microsoft Power Platform
The plugins, skills, and MCP servers that turn GitHub Copilot into a practical development partner for Dataverse, Dynamics 365, Power Apps, Power Automate, and Copilot Studio.
The coding assistant that reasons, edits files, and calls tools. Use an agent mode that supports tools.
Reusable instructions, scripts, and reference material that teach the agent a specific workflow.
Expose tools and data so the agent can interact with a service instead of just describing it.
Installable bundles of skills, agents, and tool integrations. Start with trusted, relevant bundles.
Prepare the tools and environment before the demo. Installing a plugin does not grant service permissions.
Install VS Code and GitHub Copilot. Use Agent mode.
@agentPlugins in Extensions and discovers CLI-installed plugins.Install Power Platform Tools / PAC CLI. Make sure pac is on the terminal's PATH.
Use a development or sandbox environment with synthetic data.
/plugin marketplace add microsoft/power-platform-skills
/plugin install dataverse@power-platform-skills
/plugin install model-apps@power-platform-skills
/plugin install canvas-apps@power-platform-skills
/plugin install power-automate@power-platform-skills
These are chat slash commands, not PowerShell commands. Choose only the plugins you plan to demonstrate.
The public Microsoft Power Platform Skills marketplace is the main hub. Microsoft-maintained source does not imply every feature is a fully supported product.
Connect to an environment, inspect and query data, import records, create tables and columns, and manage solutions. The plugin guides the use of MCP, the Dataverse CLI, Python SDK, and PAC CLI.
dv-connectdv-querydv-datadv-metadatadv-solutionBuild and edit tables, relationships, forms, views, charts, security roles, app navigation, and generative pages. A natural fit for Dataverse-backed business applications.
model-appsapp-buildergenpageEdit screens, controls, Power Fx, connectors, and data sources through Canvas Authoring MCP. Begin with a saved app, enable coauthoring, and keep its Power Apps Studio browser tab open.
canvas-appscanvas-appconfigure-canvas-mcpCreate and edit cloud flows, inspect failed runs, validate changes, and manage connections. Includes the FlowAgent MCP server and uses your signed-in Azure CLI identity.
create-flowbuild-flowdebug-flowdiagnose-flowmanage-flowsReact/TypeScript applications connected to Power Platform.
code-apps-previewGitHub Copilot quickstart ↗Scaffold, build, diagnose, and deploy custom controls.
pcfPCF plugin ↗Build and deploy code sites using modern web frameworks.
power-pagesPower Pages plugin ↗Know the boundaries. The Canvas Apps plugin cannot create the initial app record or enable coauthoring. Power Automate desktop-flow support covers discovery and execution, not editing Power Automate Desktop scripts.
There are two authoring plugins for different agent architectures. Choose deliberately rather than installing both by default.
Skills for Copilot Studio supports YAML authoring, validation, testing, and troubleshooting. Pair it with the Copilot Studio VS Code extension for clone, pull, and push.
/plugin marketplace add microsoft/skills-for-copilot-studio
/plugin install copilot-studio@skills-for-copilot-studioThe repository explicitly describes this as an experimental research project, not an officially supported Microsoft product.
Standard-agent plugin ↗Copilot Studio Authoring Plugin supports the newer GitHub Copilot-harness architecture, including editing, validation, and migration from classic agents. Requires PAC CLI 2.9.3+.
/plugin marketplace add microsoft/copilot-studio-plugin
/plugin install mcs-assistant@copilot-studio-pluginThe repository describes this as work in progress, with support through GitHub Issues. Review and validate generated YAML before pushing.
New-architecture plugin ↗MCP tools act with the access granted to the connected identity. Start with read-only exploration and inspect available capabilities after connecting.
| Server | Use it for | How to connect |
|---|---|---|
| Microsoft LearnRecommended for every setup | Current Microsoft documentation and code examples. No authentication required. | HTTP endpoint:https://learn.microsoft.com/api/mcpCLI plugin: /plugin install microsoftdocs/mcp |
| Dataverse MCPCore live-data connection | Inspect schemas, query data, and perform supported record and metadata operations. | Append /api/mcp to your actual environment URL. Ask the Dataverse plugin to connect. The environment must allow the Microsoft GitHub Copilot MCP client. |
| Copilot Studio MCPDirect service access | Inspect and manage supported Copilot Studio resources through the connected service's tools. | In VS Code Extensions, search @mcp copilot-studio. Package: @microsoft/copilot-studio-mcp. Sign in, choose an environment, then refresh tool discovery if needed. |
| Playwright MCPOptional / browser testing | Inspect a live app, interact with its UI, and help generate browser tests. | Follow the repository's client setup for @playwright/mcp. Use the Power Platform testing walkthrough for app-specific examples. |
Two servers are already bundled: Canvas Authoring MCP comes through canvas-apps; FlowAgent MCP comes through power-automate. Power Automate also includes Microsoft Learn MCP, so check existing registrations before adding a duplicate.
A naming trap: the separate Power Apps MCP Server is for app tasks and human-in-the-loop agent supervision. It is not Canvas Authoring MCP.
Build around customer service requests. Run these prompts in sequence, reviewing each proposed change before moving on.
Confirm the connected tenant and environment. Show the tables and relationships relevant to customer service requests. Explain what already exists. Don't change anything.
Demonstrates: grounded discovery, not guessed schemas.
Propose a service-request data model in our demo solution. Confirm the publisher and solution first. After approval, build a model-driven app with useful forms and views.
Demonstrates: solution-aware application development.
Create a cloud flow that notifies our test Teams channel when a high-priority request is created. Show the trigger, actions, and connections. Keep it disabled until I review it.
Demonstrates: real automation with an approval boundary.
Update this Copilot Studio agent to explain request status using our approved data source. Show the changes and test scenarios. Don't publish the agent.
Demonstrates: agent authoring, review, and testing.
Canvas alternative: prepare a saved blank canvas app with coauthoring enabled, connect it with configure-canvas-mcp, and ask for a request gallery and detail screen instead of the model-driven app.
Prove the workflow in development before giving the agent access to production resources.
Useful repo guidance: document the target solution, publisher prefix, approval boundaries, and validation steps in .github\copilot-instructions.md. Don't store credentials there.
The Dataverse and model-driven app tools are the closest match. Product-specific business logic and permissions still apply.
The current Dataverse skills repository includes erp-xpp for linked Finance and Operations development. It requires its own compatible environment and development setup.
AL development is a separate stack. Don't assume the Dataverse tooling covers Business Central extensions.
Installing Copilot, a skill, or an MCP server does not include every Dynamics product or bypass Power Platform licensing, billing, or tenant governance.
Public source repositories and first-party guidance. Check the current README for prerequisites, limitations, and updates.