Azure Skills Plugin: packaged Azure expertise and MCP tools for coding agents

Three layers packaged in one plugin: skills, MCP Server and Foundry MCP

The plugin ships as three distinct components that work together. The skills layer contains curated Azure expertise: workflows, decision trees and guardrails that tell an agent when to use Azure services and what to avoid. The Azure MCP Server provides the execution layer: over 200 structured tools across more than 40 Azure services that let the agent list resources, check prices, query logs and diagnose issues against live Azure data. The Foundry MCP component adds Microsoft Foundry scenarios: model discovery, model deployment, agent workflows and evaluations.

The key design decision is that the guidance layer and the execution layer are packaged together. A skill that teaches an agent when to deploy to AKS is aligned with the MCP tools that can actually perform that deployment, because both come from the same install. The README frames this as the difference between a prompt pack and a capability layer.

The plugin payload lives in the .github/plugins/azure-skills/ directory of the repository. The MCP configuration is included and shows how Azure and Foundry connectivity are wired for compatible hosts.

Supported hosts are GitHub Copilot (in VS Code and CLI), Claude Code, Cursor, OpenCode, Codex CLI, Gemini CLI and IntelliJ IDEA with the GitHub Copilot plugin. Each has its own install path, described in the sections below.

Node.js 18+, Azure CLI and Azure Developer CLI are required before install

Three prerequisites must be in place before the plugin can start. Node.js 18 or newer must be available on the PATH because npx is used to start the MCP servers. The Azure CLI must be installed and authenticated with az login. The Azure Developer CLI must be installed and authenticated with azd auth login when deployment workflows are planned.

For VS Code, the README adds a fourth prerequisite: Git CLI must be installed on the machine. Without Git, the skills extension will not load. VS Code users who need Git can ask Copilot itself to help install it.

For IntelliJ IDEA, two prerequisites must be verified before installing the GitHub Copilot plugin that the Azure skills require. Running the following two commands confirms both Node.js and Git are accessible from the command line.

npx --version
git --version

These prerequisite requirements set a minimum baseline that eliminates environments running without a Node.js runtime or without an active Azure authentication. A developer on an air-gapped system or an environment that cannot run npx cannot use the MCP server component at all.

APM installs the plugin across six hosts with one command

APM is the multi-harness path the README calls out first. One command installs the plugin across GitHub Copilot, Claude Code, Cursor, OpenCode, Codex and Gemini simultaneously from a single apm.yml configuration.

apm install microsoft/azure-skills

For teams that use multiple AI coding tools, this is the install path that avoids having to run host-specific commands for each one. APM is a separate tool maintained at github.com/microsoft/apm.

For Gemini CLI without APM, the install is a direct URL reference.

gemini extensions install https://github.com/microsoft/azure-skills

For Codex CLI, a marketplace registration step comes first: codex plugin marketplace add microsoft/azure-skills, after which the plugin can be found with /plugins and installed interactively. Codex also offers per-skill enable and disable via /skills inside the Codex interface, which means a team can install the full plugin but activate only the skills relevant to their current work.

Claude Code and GitHub Copilot use different package identifiers for the same plugin

A detail worth checking before installing: the package identifier for this plugin differs between hosts. In Claude Code, the install command is as follows.

/plugin install azure@claude-plugins-official

In GitHub Copilot CLI, the marketplace registration uses the identifier microsoft/azure-skills, but the install command uses a different name.

/plugin install azure@azure-skills

Those are two distinct package namespaces for the same plugin. A developer who sees azure@azure-skills in a GitHub Copilot guide and tries it in Claude Code will get a package-not-found error. The update commands follow the same pattern: /plugin update azure@claude-plugins-official in Claude Code, /plugin update azure@azure-skills in GitHub Copilot.

For VS Code, the install path is different again: the Azure MCP extension from the Visual Studio Marketplace is the entry point, and it installs a companion extension that brings the skills layer in. The extension ID on the Marketplace is ms-azuretools.vscode-azure-mcp-server. That extension sets up Azure MCP Server, Foundry MCP and the full skills layer automatically without a separate command.

25 named skills across four operational categories

Skills are grouped into 4 operational categories. The build, deploy and evolve category contains: azure-prepare, azure-validate, azure-deploy, azure-upgrade, azure-enterprise-infra-planner, azure-hosted-copilot-sdk, azure-kubernetes and airunway-aks-setup.

For troubleshooting, monitoring and governance, the category covers: azure-diagnostics, appinsights-instrumentation, azure-compliance, azure-resource-lookup and azure-quotas.

Architecture and cost optimization holds: azure-cost, azure-compute, azure-resource-visualizer and azure-cloud-migrate.

Data, AI, identity and platform services uses: azure-ai, azure-aigateway, azure-storage, azure-kusto, azure-messaging, azure-rbac, entra-app-registration and microsoft-foundry.

Each skill is a workflow and guardrail package rather than a single tool call. Skills teach the agent when to use Azure workflows and what to avoid, not just how to execute a command. That distinction matters for reliability: an agent acting on live Azure resources needs to know which operations are reversible and which require prior validation.

IntelliJ IDEA requires GitHub Copilot plugin version 1.5.64-242 or higher

IntelliJ IDEA support is the most conditional of all the host integrations documented in the README. The skills layer in IntelliJ runs through the GitHub Copilot plugin for JetBrains, and that plugin must be version 1.5.64-242 or higher. Earlier versions of the Copilot plugin do not support the skills mechanism.

The IntelliJ install sequence has two explicit steps: first install the GitHub Copilot plugin from the JetBrains Marketplace (plugin ID 17718), restarting IntelliJ when prompted, then enable skills for GitHub Copilot. The README's instructions for the second step are not completed; the documentation breaks off after the first substep of enabling skills.

For the Cursor host, the install can be done from the Cursor Marketplace at cursor.com/marketplace/azure or directly from Settings > Plugins inside Cursor. That path does not require a version constraint on the underlying tool.

Available under MIT, the latest release is v1.2.49, published on 2026-09-16.

When the Azure Skills Plugin is the wrong choice

The plugin is scoped to Azure. Nothing in the skills list covers AWS, Google Cloud or on-premises infrastructure. A team that works across cloud providers or has its critical workloads on a non-Azure platform gets no benefit from the Azure-specific workflows, and the MCP tools connect only to Azure and Microsoft Foundry endpoints.

The Node.js 18+ requirement also excludes environments that do not or cannot run a Node.js runtime. The MCP servers start via npx, so environments restricted from installing npm packages, environments with strict outbound network controls, or corporate machines where Node.js is not approved cannot use the execution layer. The skills files might still be readable by an agent as context, but the MCP tool calls would fail.

For teams that need a similar capability pattern on a different cloud, there is no direct equivalent in the same architecture: a cloud-agnostic skills plugin would require building or finding skills and MCP server configurations separately. The Microsoft APM tool can install third-party plugins that cover other scenarios, but those are independent of this repository.

For Azure-focused teams that want the skills layer alone without MCP, the plugin does not offer a way to install only the skills without also wiring in the MCP servers. The three components install together.

Editorial conclusion

Azure Skills Plugin is the right choice for engineers working with Azure who want their AI coding agent to take actions on live Azure resources, not just summarize documentation. It works only with Azure, so teams on AWS, GCP or mixed-cloud workloads get nothing from it. IntelliJ IDEA users need GitHub Copilot plugin version 1.5.64-242 or higher for the skills layer to load. The fastest path to verifying the install is running apm install microsoft/azure-skills if you use APM, then checking that az login and npx --version both succeed before starting an Azure workflow.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. microsoft/azure-skills on GitHub
4. README
5. Releases