Part 2 of the Azure Skills Plugin series Previously: Announcing the Azure Skills Plugin

This post is all about getting you up and running. I won’t go deep on capabilities, architecture or anything like that here – that’s coming in future posts in this series. The goal is to get the plugin installed, verified, and ready so you can start exploring right away.

The Azure Skills Plugin works across many agent hosts including GitHub Copilot CLI, VS Code, and Claude Code. Each install takes under 60 seconds. This post covers all three, what gets installed, and how to verify it’s working.

👉 Plugin repo: aka.ms/azure-skills

Prerequisites (All Hosts)

• Node.js 18+ – required for MCP servers (they run via npx)
• An Azure account – free tier works for exploring; real subscription for deployments
• Azure CLI (az) installed and authenticated (az login)
• Azure Developer CLI (azd) installed and authenticated (azd auth login)

Part 1: GitHub Copilot CLI

Install (inside Copilot CLI)

# Add the marketplace source (first time only)
/plugin marketplace add microsoft/azure-skills

# Install the plugin
/plugin install azure@azure-skills

Update

/plugin update azure@azure-skills

After Install: Reload MCP Servers

Important – Copilot CLI does not auto-load the new MCP servers. You must run:

/mcp reload

This picks up the MCP server definitions from the plugin and makes the Azure tools available.

Verify

/mcp status

You should see the MCP servers (azure, context7) listed and running.

Smoke Test

Try these prompts in Copilot CLI:

“List my Azure resource groups”

If the Azure MCP Server loaded correctly, Copilot will call the resource group tool and return live results.

“What Azure services would I need to deploy a Python web API?”

If skills loaded, you’ll see Copilot reference azure-prepare – not just generic suggestions.

Part 2: VS Code

Additional Prerequisites

• Git CLI: required on your machine for the extension to function
• VS Code Stable or Insiders: both work

Install

Install the Azure MCP extension from the VS Code Marketplace:

• 🛒 View on the VS Code Marketplace
• ⚡ Install in VS Code · Install in VS Code Insiders
• Or search “Azure MCP” in the Extensions panel (Extension ID: ms-azuretools.vscode-azure-mcp-server)

The extension auto-installs a companion extension, GitHub Copilot for Azure, which contains the Azure skills.

Verify

1. Open Copilot Chat (Ctrl+Shift+I / Cmd+Shift+I)
2. Make sure you’re in Agent mode (not Ask or Edit mode)
3. Open the Command Palette (Ctrl+Shift+P) → search “MCP” → verify servers are listed and running

Smoke Test

“List my Azure resource groups”

Copilot should call mcp_azure_mcp_group_list and return your actual resource groups.

“What AI models are available in Microsoft Foundry?”

The Azure MCP Server includes Foundry tools, so Copilot will query the model catalog directly.

Part 3: Claude Code

Install (from terminal)

# Add the marketplace source (first time only)
claude plugin marketplace add microsoft/azure-skills

# Install the plugin
claude plugin install azure@azure-skills

Update

# Update the marketplace
claude plugin marketplace update azure-skills

# Update the plugin
claude plugin update azure

What Gets Installed

All three hosts install the same plugin payload. Here’s what’s inside:

Skills → ../azure-skills/skills/ (for Copilot)

The plugin drops 21 curated skill files. Each is a markdown file (SKILL.md) that teaches the AI domain expertise for a specific Azure scenario:

.copilot/installed-plugins/azure-skills/azure/skills/
├── appinsights-instrumentation/SKILL.md
├── azure-ai/SKILL.md
├── azure-aigateway/SKILL.md
├── azure-cloud-migrate/SKILL.md
├── azure-compliance/SKILL.md
├── azure-compute/SKILL.md
├── azure-cost-optimization/SKILL.md
├── azure-deploy/SKILL.md
├── azure-diagnostics/SKILL.md
├── azure-hosted-copilot-sdk/SKILL.md
├── azure-kusto/SKILL.md
├── azure-messaging/SKILL.md
├── azure-prepare/SKILL.md
├── azure-quotas/SKILL.md
├── azure-rbac/SKILL.md
├── azure-resource-lookup/SKILL.md
├── azure-resource-visualizer/SKILL.md
├── azure-storage/SKILL.md
├── azure-validate/SKILL.md
├── entra-app-registration/SKILL.md
└── microsoft-foundry/SKILL.md

These don’t run code. They’re instruction files that the AI reads to understand when and how to help you with Azure tasks.

MCP Servers → .copilot/installed-plugins/azure-skills/azure/.mcp.json (for Copilot)

The plugin configures two MCP (Model Context Protocol) servers:

{
  "servers": {
    "azure": {
      "command": "npx",
      "args": ["-y", "@azure/mcp@latest", "server", "start"]
    },
    "context7": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-mcp@latest"]
    }
  }
}

Where Do Files Land?

The plugin payload lives under .copilot/installed-plugins/azure-skills/azure/ in your profile or workspace.

your-profile/
├── .copilot/installed-plugins/azure-skills/azure/     ← plugin payload
│   ├── skills/                                        ← 21 SKILL.md files
│   └── .mcp.json                                      ← MCP server definitions
└── ...

Skill and Tool Budgets

Each agent host has limits on how many skills or how much skill content it will include in a request. The Azure Skills Plugin ships 21 skills plus 59 MCP tools – if you’re also running other plugins or MCP servers, you may hit these limits and some skills/tools could be silently excluded.

GitHub Copilot CLI

• Skill budget: 15,000 tokens by default. If combined skill content exceeds this, some skills will be excluded. You can increase this with an environment variable.
• Context auto-compacts when it reaches 95% of the token limit. Long sessions with many skills and tool calls can trigger compaction, which may drop earlier context.
• Run /context to see a token usage breakdown.
• Run /compact to manually compress context before it hits the limit.
• Use /skills to toggle individual skills on or off – disable skills you don’t need for your current task to free up context space.

Claude Code

• Skill budget: 2% of the context window, with a fallback floor of 16,000 characters. If combined skill content exceeds this, some skills will be excluded.
• Run /context in Claude Code – if skills are being excluded, you’ll see a warning.
• Override: Set the SLASH_COMMAND_TOOL_CHAR_BUDGET environment variable to increase the budget (in characters).

Tip: If Azure skills don’t seem to be activating, check whether you have many other plugins or MCP servers loaded. Reducing the total tool/skill count often resolves the issue.

Troubleshooting

We have a Troubleshooting Guide, but some of these common tips may help fix your issues quickly.

“Skills don’t seem to be loaded”

• Copilot CLI: Run /plugin list to confirm the plugin is installed
• VS Code: Check that .github/plugins/azure-skills/ exists; reload the window (Ctrl+Shift+P → “Reload Window”); confirm you’re in Agent mode
• Claude Code: Run /skills inside Claude Code to confirm

“MCP tools aren’t showing up”

• Copilot CLI: Run /mcp reload – this is required after install – then run /mcp show azure to see the tools
• VS Code: Check .mcp.json in workspace root; try Command Palette → “MCP: Restart Servers”; check Output panel (View → Output → “MCP”)
• Claude Code: Run /mcp and navigate to the MCP servers to see the tools
• All hosts: Make sure Node.js 18+ is installed (node --version)

“Azure commands fail with auth errors”

• Run az login to refresh authentication
• Run azd auth login for deployment skills
• Set your subscription: az account set --subscription <your-sub-id>

“I already had Azure MCP installed separately”

No problem. The plugin’s .mcp.json uses the same @azure/mcp package. You’re just getting the skills layer on top. Your existing MCP tools continue to work.

What’s Next

You’re installed and verified. In the next post, we’ll look at how skills and MCP servers actually work together – how a skill decides to activate, how it orchestrates MCP tool calls, and how progressive disclosure keeps everything fast.

🚀 Try It Yourself

1. Install: aka.ms/azure-skills
2. Open your AI coding tool (Copilot CLI, VS Code, or Claude Code)
3. Ask: “List my Azure resource groups”
4. Ask: “What would I need to deploy a Python Flask API to Azure?”

If both work, you’re ready for the rest of the series.

Previous: Meet the Azure Skills Plugin → Next up: Building with Azure Skills →