Azure Ai Agent Demos

Azure AI Foundry Agents Demos

Sample code and walkthroughs for building Azure AI Foundry Agents using the GA v2 SDK (azure-ai-projects>=2.0.0). Demos cover MCP server integration, Code Interpreter, and enterprise patterns like credential isolation via Foundry project connections.

Demos (v2 — GA SDK + New Foundry UI)

| Demo | Description | Tools | Tests | |------|-------------|-------|-------| | [enterprisegithubagent](enterprisegithubagent/) | GitHub integration with code execution | MCP (GitHub) + Code Interpreter | 42 | | [mcpmslearnagent](mcpmslearnagent/) | Search Microsoft Learn documentation | MCP (MS Learn, public) | 41 | | [mcplocalserveragent](mcplocalserveragent/) | Local MCP server with Chinook SQLite DB | MCP (custom, local + ngrok) | 84 |

Every demo follows a consistent pattern:

• create_agent.py — Create the agent (run once)
• ask_agent.py — Interactive chat REPL with streaming output
• .env.sample — Required environment variables
• requirements.txt — Python dependencies
• README.md — Setup and walkthrough

Quick Start

# Pick a demo directory
cd enterprise_github_agent  # or mcp_mslearn_agent, mcp_local_server_agent

# Set up
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.sample .env
# Fill in PROJECT_ENDPOINT and MODEL_DEPLOYMENT_NAME

# Create the agent and chat
python create_agent.py
python ask_agent.py

See each demo's README for detailed setup instructions.

Prerequisites

• Python 3.11+
• An Azure AI Foundry project with a deployed model
• Azure CLI logged in (az login) for DefaultAzureCredential

Architecture

All demos use the GA V2 SDK:

• Agent creation via PromptAgentDefinition + create_version()
• Conversations via OpenAI Responses API (responses.create())
• Credentials in Foundry project connections (not in code)

See [docs/architecture/overview.md](docs/architecture/overview.md) for details and [docs/architecture/decisions/](docs/architecture/decisions/) for ADRs.

Archived Demos (v1)

The original demos from the walkthrough video are preserved in [archive/v1/](archive/v1/). They target the deprecated v1 Agents SDK (azure-ai-agents) and the legacy Foundry portal UI.

• [v1 MCP Microsoft Learn Agent](archive/v1/mcpmslearnagent/)
• [v1 MCP Local Server Agent](archive/v1/mcplocalserver_agent/)

Roadmap

• [x] Enterprise GitHub Agent (Phase 1)
• [x] MCP Microsoft Learn Agent v2 rewrite (Phase 2)
• [x] MCP Local Server Agent v2 rewrite (Phase 3)
• [x] Cross-cutting polish — architecture docs, README (Phase 4)
• [ ] Fabric Data Agent — natural language queries against Fabric lakehouse (Phase 5)
• [ ] Multi-Tool Knowledge Worker — File Search + Code Interpreter + Web Search (Phase 6)
• [ ] Browser Automation Agent — browser control via natural language (Phase 7)

Running Tests

pip install pytest
python -m pytest enterprise_github_agent/tests/ mcp_mslearn_agent/tests/ mcp_local_server_agent/tests/ -v
# 167 tests, all structural (no live Azure calls needed)

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: jcentner
• Source: jcentner/azure-ai-agent-demos
• License: MIT

Install and usage instructions live in the source repository linked above.