From Code to Agents: Build Production MCP Servers on Azure Functions
Jonah Andersson explains how to take AI agents from prototype to production by building MCP servers on Azure Functions, using .NET to expose tools that agents can discover and invoke at runtime. The session uses a DevOps CI/CD scenario where agents detect deployment issues, diagnose failures, and create rollback PRs in Azure DevOps.
Overview
What problem this session targets
- AI agents can be useful in prototypes, but production systems often fail due to fragile, custom tool integrations.
- Agents need a reliable way to call real tools (for example, DevOps systems) without bespoke glue code that breaks when APIs or workflows change.
MCP as the tool-invocation standard
- MCP (Model Context Protocol) is presented as a standard way for agents to:
- Discover available tools at runtime.
- Invoke those tools through a consistent interface.
- The goal is to reduce one-off integrations and make agent-to-tool connectivity more maintainable.
Reference scenario: DevOps CI/CD automation with rollback PRs
The session describes a real-world DevOps automation scenario where agents:
- Identify and detect deployment issues.
- Diagnose what went wrong.
- Create rollback pull requests (PRs) in Azure DevOps.
Azure and .NET implementation approach
- Azure Functions is used as the compute layer for hosting MCP tool extensions.
- The MCP tool extension is coded in .NET.
- The session also references using:
- Azure AI Foundry
- Microsoft Agent Framework (MAF)
Session context
- Originally presented in-person at Microsoft Build 2026 San Francisco (LTG455) and recorded for the Microsoft developer community.
Speaker
- Jonah Andersson (Principal Cloud Architect at Jonah Andersson Tech; Microsoft MVP; Microsoft Certified Trainer; .NET and Azure specialist)
Links
- LinkedIn: https://www.linkedin.com/in/jonahandersson
- GitHub: https://github.com/jonahandersson
- X: https://x.com/cjkodare
- Book: Learning Microsoft Azure (O'Reilly) https://learningmicrosoftazure.com/