Azure Skills: Microsoft's Official MCP Plugin That Gives Coding Agents 200+ Azure Tools

The short version

Azure Skills addresses a very specific pain point: coding agents usually know what should be done but lack the permission and the machinery to actually do it. Microsoft packages the operational know-how of Azure (the Skills layer) together with a set of MCP tools that can perform real Azure operations (the execution layer) into a single plugin, so coding agents such as GitHub Copilot, Claude Code, Cursor and Codex gain end-to-end capability for deployment, troubleshooting, governance and cost analysis in one install.

What Azure Skills is

Azure Skills is Microsoft's official agent plugin that supplies coding agents with Azure-specific skills and MCP Server configuration. It bundles Azure domain knowledge with more than 200 structured tools, enabling compatible agents — GitHub Copilot, Claude Code, Cursor, Codex and others — to carry out genuine Azure resource management, deployment, monitoring and optimization work.

It is important to stress what it is not: this is not a prompt pack. A prompt pack only teaches the model "how to think about this kind of problem," whereas Azure Skills also provides "which Azure interface to call and how." The former produces prose in a chat window; the latter produces changes on real cloud resources.

Core capabilities

Build and deploy — Skills such as azure-prepare, azure-validate, azure-deploy and azure-upgrade cover the whole application lifecycle: pre-release environment checks, configuration validation, the actual deployment, and version upgrades. That forms a continuous loop you can drive entirely from a conversation. Instead of bouncing between docs, command-line references and the portal, you describe the goal and let the agent advance stage by stage.

Troubleshooting and governance — The plugin combines azure-diagnostics, appinsights-instrumentation and azure-compliance to support production incident diagnosis and compliance auditing. The first pulls logs and locates anomalies, the second wires Application Insights telemetry into your application, and the third scans resource configuration against best practices. That compresses an incident workflow — read logs, add monitoring, fix config, come back later for a compliance pass — into a single dialogue.

Cost and architecture optimization — azure-cost, azure-compute and azure-resource-visualizer deliver cost analysis and architecture visualization. Cost data is normally lagging and only visible after the fact; once pricing lookups and usage statistics become agent-callable tools, "how much will this architecture cost per month?" turns into something you can casually ask while writing code. Resource visualization helps you understand the current topology rather than staring at a pile of resource IDs.

Cross-service integration — Support for azure-ai, azure-storage, azure-kusto, azure-messaging, azure-rbac and 40+ other Azure services. Coverage runs from virtual machines and container services to storage, messaging and the permission model itself. That breadth is what makes this a general-purpose operation layer for the entire platform rather than a mere "deployment helper."

Foundry workflows — A built-in Foundry MCP handles model discovery, deployment and agent workflow orchestration. For teams building AI applications, this layer stitches together the model catalog, endpoint deployment and workflow orchestration, so data scientists can let the agent discover available models and stand up an endpoint instead of combing through documentation one page at a time.

How to use Azure Skills

Prepare the environment — You need an Azure subscription, Node.js 18 or newer, a signed-in Azure CLI (run az login), and a signed-in Azure Developer CLI (run azd auth login). This step is mandatory: the plugin operates on cloud resources through your local credentials, so without a session the tool calls have nothing to act on.

Pick an installation path — Either use the one-line APM install (apm install microsoft/azure-skills) or install per target platform, covering Copilot CLI, VS Code, Claude Code, Cursor, Codex, Gemini CLI and IntelliJ IDEA.

Per-platform specifics — VS Code users install the Azure MCP Extension from the Marketplace; it configures both the MCP Server and the Skills layer automatically, so you rarely edit config files by hand. Claude Code users run /plugin install azure@claude-plugins-official or simply search for "azure" in the plugin list. Codex users first run codex plugin marketplace add microsoft/azure-skills, then browse /plugins to install azure.

Enable and verify — After installation you can toggle individual skills: run /skills in Codex to enable or disable specific Azure skills. To confirm everything landed, type /skill:azure in Copilot Chat to list the available skills. Enabling everything at once adds unnecessary context overhead, so it is best to switch on only what the current task requires.

Why it stands out

• Real execution, not advice. Unlike a prompt pack, it ships operational MCP tools that manipulate Azure resources directly; the agent's output is something that happened, not a paragraph of suggestions.
• Three aligned layers. The Skills guidance layer and the MCP execution layer ship inside the same plugin, closing the gap between knowing what to do and being able to do it.
• Multi-host coverage. One plugin serves Copilot, Claude Code, Cursor, Codex, Gemini and IntelliJ, so teammates on different editors can still collaborate.
• Breadth: 200+ tools. Spanning 40+ Azure services — from virtual machines to AI services, from cost monitoring to compliance auditing — all under one namespace.
• Sovereign cloud support. Configurable for Azure China Cloud, Azure US Government and similar environments, satisfying data residency and regulatory requirements.

Where it fits

• Cloud-native development — Ask the agent to create an AKS cluster, configure an Azure AI Gateway or deploy the application mid-coding without breaking your flow to open a portal.
• Cost governance — Finance or platform teams query historical spend, forecast future cost and surface waste from idle or misconfigured resources.
• Compliance auditing — Security teams run azure-compliance to automatically check for expired Key Vault secrets, over-privileged RBAC assignments and best-practice drift.
• Cross-cloud migration — Evaluate migration paths and refactor points when moving workloads such as AWS Lambda to Azure Functions.
• AI infrastructure — Data scientists rely on Foundry MCP to discover models, deploy endpoints and manage agent workflows, automating the model release pipeline.

Repository and open-source status

The project is maintained in the open at https://github.com/microsoft/azure-skills by Microsoft's own team. The skill inventory, changelog and contribution guide are all public, so enterprise users can complete an internal security review before adopting it.

How it compares to similar tools

Comparing it with browser-act, another skill-shaped tool, makes the difference in positioning clear. Azure Skills is a cloud-infrastructure agent plugin aimed at Azure resource deployment, monitoring, cost governance and compliance auditing; browser-act is a browser-automation skill library for page browsing, DOM manipulation, form filling and data scraping.

On protocol, Azure Skills builds on MCP (Model Context Protocol), an open standard, while browser-act uses a private skill registry. Their execution targets differ too: the former addresses Azure REST APIs, the Azure CLI and the resource graph; the latter addresses browser engines and DOM APIs. In scale, the former offers 200+ structured tools covering resource inventory, pricing queries, log diagnostics and RBAC configuration; the latter offers roughly 15–20 browser actions such as click, type, scroll, screenshot and navigate.

Agent binding differs as well: Azure Skills supports multiple harnesses and can be reused across Copilot, Claude, Cursor, Codex, Gemini and IntelliJ after a single install, whereas browser-act typically binds to Claude or one specific agent environment. At the decision layer, Azure Skills provides curated workflows, decision trees and guardrails — telling you when to deploy and which risks to avoid — while browser-act orchestrates action sequences that execute browsing tasks step by step. For safety boundaries, Azure Skills leans on Azure RBAC, compliance guardrails and cost alerts; browser-act relies on browser sandboxing and agent permission controls.

In short, the two are complementary rather than substitutable: one manages resources in the cloud, the other manages pages in a browser, and real projects often need both.