Agent
An agent definition bundles the model, system prompt, tools, MCP servers, and skills you want to reuse.
What Is It
What is Claude Managed Agents?
Use this page when you need a compact explanation of the hosted runtime, the core concepts, and the workloads that match Managed Agents best.
Last updated
2026-04-12
Core Concepts
Managed Agents is easier to understand once you stop thinking in one giant “agent” blob.
Anthropic’s overview keeps the product grounded in four primitives. That mental model is enough to start.
Environment
An environment defines the hosted container template, including packages, files, and network access rules.
Session
A session is the running instance that performs a task, streams events, and keeps a durable event history.
Events
Events are the messages and tool updates your application sends and receives while the session works.
Responsibility Split
The product value is not just “Claude can use tools.” It is that the hosted runtime is already there.
This split is the fastest way to understand where Managed Agents ends and your application begins.
| Dimension | What you own | What Anthropic owns |
|---|---|---|
| Build and runtime responsibilities | You define the agent configuration, environment settings, session creation, user events, and how your app consumes outputs. | Anthropic provides the hosted harness, session model, event persistence, and managed container runtime. |
| Operational responsibilities | You decide the task, guardrails, custom tools, and product UX around the agent. | Anthropic handles the underlying service that runs the agent loop, tool execution flow, and session lifecycle primitives. |
When It Fits
It earns its complexity when the runtime and session model matter.
Anthropic explicitly recommends Managed Agents for long-running tasks, cloud infrastructure, minimal infrastructure burden, and stateful sessions.
Use it for long-running work that benefits from built-in tool access and a durable session log.
Use it when you want Anthropic-managed cloud infrastructure instead of building your own harness and sandbox.
Do not use it just because the task is “agentic.” If you only need a short request-response loop, start smaller.
Do not confuse it with Claude Code if the real user is a human developer in an IDE or terminal.
Why Anthropic Built It
Managed Agents is meant to give you a durable interface around evolving agent infrastructure.
This explains why the product exists instead of simply telling you that Claude can already use tools.
The engineering post is helpful here: Anthropic describes Managed Agents as a service designed to keep the interfaces around Claude stable even as the underlying harness changes. That is why the docs focus on sessions, sandboxes, and events instead of exposing every implementation detail.
FAQ
The cleanest definition is hosted agent infrastructure for long-running work.
These questions help reset expectations before a team jumps into implementation.
Is Managed Agents just another Messages API mode?+
No. Anthropic treats the Messages API and Managed Agents as two different build paths: one gives you direct prompting access, the other gives you a pre-built configurable agent harness on managed infrastructure.
Is every Managed Agents feature generally available?+
Not by default. Anthropic documents access as enabled by default for API accounts, but features such as outcomes, multiagent, and memory are still research preview.
Related Pages
Keep moving with the page that matches the next decision.
These pages are the shortest path to the next question most teams hit.
Claude Managed Agents quickstart
Use this guide to build the first working Managed Agents session without mixing up agents, environments, sessions, and event streams.
Open pageClaude Managed Agents vs Agent SDK
Use this comparison when the real choice is between Anthropic-managed infrastructure and a programmable SDK stack you operate yourself.
Open pageClaude Managed Agents vs Claude Code
Use this page to separate a hosted agent service from the developer-facing Claude Code product.
Open pageStay Updated
Want updates as Managed Agents evolves?
Use the waitlist if you want the next update without checking the docs every week.
Waitlist
Want updates as Managed Agents evolves?
Get a short email when this guide adds new examples, API changes, or managed-agents implementation notes.