Microsoft Azure AI Foundry Skills - Setup Guide¶
Source: microsoft/azure-skills (478K+ combined installs) Category: Agent Infrastructure / Cloud Quality Tier: 🟢 Production
Microsoft's official skills for building, deploying, and managing AI agents on Azure AI Foundry. Covers the full agent lifecycle: project creation, model deployment, agent scaffolding, CI/CD pipelines, observability, evaluation, fine-tuning, and troubleshooting. Enterprise-grade infrastructure for production agent deployments.
Installation¶
npx skills add microsoft/azure-skills --skill microsoft-foundry
npx skills add microsoft/azure-skills --skill azure-ai
npx skills add microsoft/azure-skills --skill azure-deploy
Included Skills¶
| Skill | Installs | Purpose |
|---|---|---|
| microsoft-foundry | 478.2K | Full Azure AI Foundry agent platform - create, deploy, invoke, observe, CI/CD, routines, fine-tuning |
| azure-ai | 474.4K | Azure AI Services resource and project management |
| azure-deploy | 474.1K | Agent deployment, versioning, and multi-environment management |
| azure-hosted-copilot-sdk | 443.4K | Hosted Copilot SDK for building agents on Azure AI Foundry (added Sep 29, 2026 sweep) |
Prerequisites¶
| Requirement | Details |
|---|---|
| Azure subscription | Active Azure account with AI Foundry access |
Azure CLI (az) |
Install via curl -sL https://aka.ms/InstallAzureCLIDeb |
Azure Developer CLI (azd) |
Install via curl -fsSL https://aka.ms/install-azd.sh |
| Python 3.10+ | For hosted agent development |
Key Capabilities¶
Agent Lifecycle Management¶
- Create: Scaffold new agents end-to-end (quick-start) or customize with existing code, A2A, and advanced setups
- Deploy: Deploy hosted agents to Foundry, manage versions, multi-environment deploys
- Invoke: Send messages to agents, single or multi-turn conversations, WebSocket duplex for voice/real-time
- Routines: Schedule or event-trigger agents with CRUD, enable/disable, manual dispatch
Observability & Quality¶
- Observe: Evaluate agent quality, run batch evals, analyze failures, optimize prompts
- Trace: Query traces, analyze latency/failures, correlate eval results
- Agent Optimizer: Run optimization jobs, apply candidates locally, deploy through azd
- Eval Datasets: Harvest production traces into datasets, version management, regression detection
Infrastructure¶
- Project/Resource Creation: Create Foundry projects with public or VNet-isolated access
- Model Deployment: Unified deployment with intelligent routing (preset, custom, capacity discovery)
- Fine-tuning: SFT distillation, DPO preference optimization, RFT with graders
- CI/CD: Set up deployment pipelines with azd, continuous evaluation monitoring
Quick Start¶
# 1. Authenticate
az login
# 2. Create a new Foundry project (public access)
azd init --template microsoft/azure-ai-foundry-agent
azd up
# 3. Deploy and test an agent
azd deploy
azd invoke --agent my-agent --message "Hello"
# 4. Set up CI/CD
azd pipeline config
Verification¶
npx skills list | grep microsoft/azure-skills
az account show # Verify Azure auth
azd version # Verify azd CLI
Notes¶
- Official Microsoft-maintained skills with 478K+ combined installs - the standard for Azure AI agent infrastructure
- Requires active Azure subscription and cloud resources (not local-only)
- Best for production-grade agent deployments with enterprise requirements (VNet, CI/CD, observability)
- Complements local agent development (Hermes, Claude Code) with cloud deployment capabilities
- Fine-tuning supports SFT, DPO, and RFT with custom graders
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.