Welcome to the Foundry Hosted Agents Workshop — a hands-on, progressive workshop built directly from a real proof of concept: hosting a LangGraph multi-agent threat-assessment system on Microsoft Foundry Hosted Agents, backed by independent MCP tool servers running on Azure Container Apps.
You will deploy the MCP tool servers, provision a Foundry project and a hosted agent, invoke it, gate it with deterministic and LLM-as-judge evaluations, walk the CI/CD pipeline that promotes a candidate to production, and study the resolved WI-11 investigation, including the difference between observed recovery and an unconfirmed platform root cause.
[!NOTE] This workshop is built from the
foundry-hosted-agentsrepository. Deployment and evaluation evidence are real; MCP security telemetry uses synthetic fixtures, not live customer data. Historical investigation excerpts are retained and labeled separately from current results.
Current Pilot Links
Open the staging web chatbot with an approved pilot account. The UI uses synthetic security data and targets staging, not production. Health check.
The September 16 hybrid release passed with production version 37; see the current validation record. The September 8 artifacts below are historical evidence.
Verified Release: September 8, 2026
Run 34178081808 passed all seven release jobs: staging version 6, eight captured cases, 21/21 model-judge checks, normal production approvals, production version 34, and its exact-version smoke test. WI-11 is operationally resolved.
See the release evidence and operations runbook. The images are artifact renderings, not portal screenshots. Recovery remains manual; zero exceptions in a trailing window is not an endurance test or proof of complete distributed tracing.
Who Is This For?
| Audience | What You Will Learn |
|---|---|
| AI / platform engineers | Deploy a multi-agent LangGraph system on Foundry Hosted Agents end to end |
| DevOps engineers | Wire evaluation-gated CI/CD pipelines around an agent deployment |
| Solution architects | Compare Foundry Hosted Agents against self-hosted LangGraph/LangSmith options |
| Support / SRE engineers | Learn a methodical Azure RBAC/permission-troubleshooting workflow |
Prerequisites
Before starting Lab 00, ensure you have the following:
- Visual Studio Code (latest stable)
- Python 3.13
- Azure Developer CLI (
azd) - Azure CLI (
az) - An Azure subscription with access to Microsoft Foundry and
gpt-4o-minimodel quota - A GitHub account with GitHub Copilot access (optional, used in Lab 01)
Labs
| # | Lab | Duration | Level |
|---|---|---|---|
| 00 | Prerequisites and Environment Setup | 20 min | Beginner |
| 01 | Architecture Deep Dive | 30 min | Beginner |
| 02 | Deploy the MCP Tool Servers | 30 min | Intermediate |
| 03 | Provision and Deploy the Hosted Agent | 35 min | Intermediate |
| 04 | Invoke the Agent and Read Traces | 30 min | Intermediate |
| 05 | Evaluations: Deterministic + LLM-as-Judge | 40 min | Intermediate |
| 06 | CI/CD: Evaluation-Gated Release Pipeline | 35 min | Advanced |
| 07 | Real-World Troubleshooting: RBAC 401 | 40 min | Advanced |
| 08 | Production Readiness and Decision Gates | 30 min | Advanced |
| 09 | Teardown and Cost Cleanup | 10-20 min | Beginner |
Workshop Schedule
Half-Day (3 hours)
| Time | Activity |
|---|---|
| 0:00 – 0:20 | Lab 00: Prerequisites |
| 0:20 – 0:50 | Lab 01: Architecture Deep Dive |
| 0:50 – 1:20 | Lab 02: Deploy the MCP Tool Servers |
| 1:20 – 1:55 | Lab 03: Provision and Deploy the Hosted Agent |
| 1:55 – 2:10 | Break |
| 2:10 – 2:40 | Lab 04: Invoke the Agent and Read Traces |
| 2:40 – 3:00 | Lab 05: Evaluations (start) |
Full-Day (6 hours)
| Time | Activity |
|---|---|
| 0:00 – 3:00 | Half-Day labs (as above) |
| 3:00 – 3:15 | Break |
| 3:15 – 3:40 | Lab 05: Evaluations (continued) |
| 3:40 – 4:15 | Lab 06: CI/CD Pipeline |
| 4:15 – 4:55 | Lab 07: Real-World Troubleshooting: RBAC 401 |
| 4:55 – 5:10 | Break |
| 5:10 – 5:40 | Lab 08: Production Readiness and Decision Gates |
| 5:40 – 6:00 | Lab 09: Teardown, verification and Q&A |
Delivery Tiers
| Tier | Labs | Duration | Audience |
|---|---|---|---|
| Half-Day | Labs 00 – 05 (start), then 09 | ~3 hours + cleanup | First exposure to Foundry Hosted Agents |
| Full-Day | Labs 00 – 09 | ~6 hours | Deployment, evaluation, local CI gates, troubleshooting and cleanup |
Times are planning estimates, not provisioning guarantees. Allow extra time for tool installation, image builds, quota or identity propagation. Reserve cleanup time even when an earlier step fails.
Getting Started
- Clone or fork the
foundry-hosted-agentsrepository. - Complete Lab 00: Prerequisites to set up your environment.
- Work through the labs in order — each lab builds on the previous one.
[!IMPORTANT] Labs 02-04 create resources in your approved disposable learner group. Do not use historical customer endpoints or shared GitHub environments. Complete Lab 09 before leaving, even if deployment failed.
Slide Deck
A companion slide deck (English and French) is available for instructor-led delivery:
Related Resources
| Resource | Description |
|---|---|
foundry-hosted-agents repository |
Full PoC source code, infra, evaluation suite, and CI/CD pipelines |
| Project wiki | Architecture notes, manual-agent workaround, RBAC 401 investigation log |
| Decision deliverables | Production decision-gate scorecard and executive decision deck |
License
This project is licensed under the MIT License.