AI coding agents are good at writing code. They’re not good at knowing which Azure service fits your workload, which SKU makes sense, what needs to be validated before deployment, or which permissions and quotas matter. That gap between writing code and getting it to production is exactly what Microsoft’s new Azure Skills Plugin is designed to close.
Announced March 9 by Chris Harris on the All Things Azure blog, the plugin bundles 19+ curated Azure skills, the Azure MCP Server with over 200 tools across 40+ services, and the Foundry MCP Server for AI model workflows — all in a single install. It works across GitHub Copilot in VS Code, Copilot CLI, Claude Code, and other tools that support the agent plugin and skills patterns.
The timing isn’t accidental. This is one of the first major plugins built on the VS Code agent plugin architecture that shipped in VS Code 1.110 just days earlier. And it demonstrates what that architecture looks like when a cloud platform vendor fills it with real operational knowledge.
Three Layers in One Install
The plugin packages three distinct capabilities that work together.
Skills are the decision layer. The 19+ Azure skills package expert knowledge as reusable workflows, decision trees, and guardrails. They don’t just list tools — they tell the agent when to use them, in what order, and what to avoid. Key skills include azure-prepare (analyzes your project and generates infrastructure code, Dockerfiles, and deployment config), azure-validate (runs pre-flight checks before deployment), azure-deploy (orchestrates the actual deployment pipeline), azure-cost-optimization (finds waste and produces savings recommendations), and azure-diagnostics (troubleshoots failures with logs, metrics, and KQL queries). Additional skills cover compute, observability, compliance, storage, migration, RBAC, and messaging.
The Azure MCP Server is the execution layer. It provides over 200 structured tools across 40+ Azure services — listing resources, checking prices, querying logs, running diagnostics, provisioning infrastructure, and driving deployment workflows. The agent doesn’t just advise. It acts.
The Foundry MCP Server is the AI specialist. It connects the agent to Microsoft Foundry for model deployment, agent management, and model catalog workflows. For teams building AI-powered applications on Azure, this layer handles model and agent scenarios that generic cloud guidance can’t.
The design principle: skills guide, MCP executes, the plugin packages both so the guidance and execution layers stay aligned.
What Changes for Developers
The before-and-after comparison in the announcement is telling. Ask an agent without the plugin to deploy a Python Flask API to Azure, and you get a generic tutorial — maybe some az commands, maybe a docs link. Ask the same agent with the plugin, and it activates azure-prepare, azure-validate, and azure-deploy in sequence, producing a Dockerfile, infrastructure files, deployment config, validation output, and real MCP tool calls against Azure.
Same prompt. The difference is that the agent has a playbook and the tools to follow it, rather than generating advice.
Skills load on demand — the agent reads their name and description to decide which are relevant, then loads only what’s needed. You can install the full skill set without cluttering every prompt. They’re plain text, version-controlled, and auditable. One team can package hard-won Azure knowledge once and reuse it across projects and engineers.
According to Mitch Ashley, VP and practice lead for software lifecycle engineering at The Futurum Group, “Microsoft is packaging institutional cloud knowledge as versioned, executable skills that agents consume automatically, collapsing the gap between writing code and deploying it to production. Skills encode the decision logic, MCP tools execute against real resources, and the plugin architecture keeps guidance and execution aligned. That’s agent orchestration with guardrails, beyond just automation with documentation.”
Ashley continues, “The tribal knowledge that once lived in a senior engineer’s head becomes an installable, reviewable artifact. Teams that figure out how to package their own expertise in this model will accelerate deployment cycles. Those waiting for agents to infer that knowledge from generic training will keep getting generic answers.”
Why This Matters for DevOps
The Azure Skills Plugin is a concrete example of a pattern that’s becoming central to how DevOps teams work with AI agents. We’ve covered agent plugins as infrastructure, IronCurtain’s policy-as-code model, Google ADK’s integration ecosystem, and Copilot’s Jira bridge. The Azure Skills Plugin sits at the intersection of all of them: Packaged expertise, delivered through a plugin architecture, with both guidance and execution capabilities built in.
For platform engineering teams, the plugin represents a model worth watching. Instead of writing documentation that developers may or may not read, you package the knowledge as skills that agents consume automatically. The agent doesn’t need to read your Confluence page about which Azure services to use. It activates the skill that encodes that decision tree.
That’s a shift in how institutional knowledge gets distributed. Documentation explains. Skills execute. The tribal knowledge that used to live in a senior engineer’s head — which SKU to pick, which pre-flight checks to run, which cost traps to avoid — becomes a versioned, reviewable, installable package.
The plugin also ships guardrails alongside capabilities. Skills include validation steps, permission checks, and cost awareness. The agent operates against real Azure resources, so curated skills, trusted plugin sources, tool approvals, and least-privilege access all matter. This isn’t simulated cloud work. It’s real provisioning, real deployment, real cost implications.
Install the Azure Skills Plugin at aka.ms/azure-plugin. One install, three layers, and a much different answer to “deploy my app to Azure.”