Microsoft Foundry Agent Service Documentation
by Microsoft
Microsoft's reference for taking an agent from prototype to a governed, identity-scoped production deployment on Azure.
Overview
Foundry Agent Service is Microsoft's managed agent runtime, and this documentation set is organized around the decision it forces first: how much you build versus how much Azure manages. Prompt agents are pure configuration — instructions, a model from the Foundry catalog, and tools — and Foundry runs them with no code or containers to maintain. Hosted agents are code-first: you ship a container image or a source zip built with Microsoft Agent Framework, LangGraph, the OpenAI Agents SDK, the Anthropic Agent SDK, the GitHub Copilot SDK or your own stack, and Foundry gives it a managed endpoint, autoscaling, session-level state and a dedicated Microsoft Entra identity, with each session running in a VM-isolated sandbox that scales to zero when idle. A third path, calling the Responses API directly, builds ephemeral agents whose definition lives entirely in your application code. The docs then cover toolboxes — curated tool sets exposed behind one managed MCP-compatible endpoint with centralized authentication, governance and versioning — built-in tools including web search, file search, code interpreter and memory, remote MCP server connection using key-based, Entra or OAuth on-behalf-of authentication, and a seven-stage development lifecycle running create, test, trace, evaluate, optimize, publish and monitor. Enterprise sections cover RBAC, private networking with bring-your-own VNet, content filters for cross-prompt injection, and publishing into Microsoft Teams and Microsoft 365 Copilot. Hosted agents reached general availability in July 2026 and the overview carries an August 2026 update stamp.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- Reference documentation; ~3-4 hours for the overview, concepts and quickstarts
- Provider
- Microsoft
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓When to choose a prompt agent, a hosted agent, or a direct Responses API call
- ✓Package an existing LangGraph or OpenAI Agents SDK agent as a hosted container
- ✓Build a toolbox that exposes curated tools behind one managed MCP-compatible endpoint
- ✓Connect remote MCP servers using key-based, Entra, or OAuth on-behalf-of authentication
- ✓Give each agent its own Microsoft Entra identity and scope its permissions with RBAC
- ✓Trace every model call and tool invocation, then evaluate agents to catch regressions
- ✓Run agents inside a virtual network and publish them to Teams and Microsoft 365 Copilot
Highlights
- •Framework-agnostic: LangGraph, Agent Framework, OpenAI and Anthropic SDKs all run as hosted agents
- •Per-agent Microsoft Entra identity is a first-class concept, not bolted on afterwards
- •Toolboxes expose a curated tool set as one managed MCP endpoint with versioning and governance
- •Hosted agents run in per-session VM-isolated sandboxes that scale to zero when idle
- •Continuously revised documentation — the overview page carries an August 2026 update stamp
Who It’s For
Best For
- ✓Enterprise teams moving an agent prototype onto governed Azure infrastructure
- ✓Platform engineers who need agent identity, RBAC and network isolation by default
- ✓Developers with an existing LangGraph or OpenAI Agents SDK codebase to host
- ✓Architects comparing a managed agent runtime against self-managed Kubernetes
Prerequisites
- •Working knowledge of Azure resources, Microsoft Entra identity and RBAC
- •Experience building an agent with at least one framework or SDK
- •Familiarity with containers, since hosted agents ship as images or source archives
FAQ
What is Microsoft Foundry Agent Service Documentation?
Microsoft's official documentation for Foundry Agent Service, the managed platform that hosts, scales and governs AI agents on Azure. It covers the two agent types — configuration-only prompt agents and containerized hosted agents built with Agent Framework, LangGraph, the OpenAI or Anthropic Agent SDKs, or your own code — plus toolboxes, remote MCP connectivity, per-agent Entra identity, tracing, evaluation and publishing. Written for engineers moving an agent prototype into governed production.
Is Microsoft Foundry Agent Service Documentation free?
Microsoft Foundry Agent Service Documentation is free to access.
What level is Microsoft Foundry Agent Service Documentation for?
Microsoft Foundry Agent Service Documentation is aimed at a intermediate audience. Recommended background: Working knowledge of Azure resources, Microsoft Entra identity and RBAC, Experience building an agent with at least one framework or SDK, Familiarity with containers, since hosted agents ship as images or source archives.
How long does Microsoft Foundry Agent Service Documentation take?
Expect roughly Reference documentation; ~3-4 hours for the overview, concepts and quickstarts. Most learners work through it at their own pace.
What will I learn from Microsoft Foundry Agent Service Documentation?
You'll learn: When to choose a prompt agent, a hosted agent, or a direct Responses API call; Package an existing LangGraph or OpenAI Agents SDK agent as a hosted container; Build a toolbox that exposes curated tools behind one managed MCP-compatible endpoint; Connect remote MCP servers using key-based, Entra, or OAuth on-behalf-of authentication; Give each agent its own Microsoft Entra identity and scope its permissions with RBAC; Trace every model call and tool invocation, then evaluate agents to catch regressions; Run agents inside a virtual network and publish them to Teams and Microsoft 365 Copilot.
Topics
Sources
This page was written from 2 sources, 1 on domains other than learn.microsoft.com.