What it is
Start here if you need the shortest explanation of what Managed Agents is, what it runs for you, and where it stops being the right tool.
Read the overviewDeveloper guide
Claude Managed Agents explained for developers.
Use this site to understand the hosted runtime, choose the right Anthropic path, start the first session, and avoid the implementation mistakes that show up when teams guess how Managed Agents works.
Last updated
2026-04-12
Waitlist
Get the next implementation update without re-checking every doc page.
Use the waitlist if you want fresh examples, troubleshooting notes, and API changes as Managed Agents evolves.
Start Here
Three entry points cover the decisions most developers need first.
If you only have ten minutes, start with one of these.
Quickstart
Follow the first-session path: create an agent, define an environment, open a session, then stream events back to your app.
Open the quickstartvs Agent SDK
Use the comparison if you are choosing between Anthropic-managed infrastructure and a custom orchestration layer you control yourself.
Compare with Agent SDKTasks
Pick the page that matches the job you need to finish.
This hub is organized around developer tasks, not around platform marketing.
Figure out what it actually is
Clarify the product shape, the core concepts, and the real boundary between Messages API, Managed Agents, and Claude Code.
Understand the modelShip the first proof of life
Use a compact first-session walkthrough when you want to move from docs to a working agent without building your own harness first.
Start buildingChoose the right Anthropic path
Use the two comparison pages when the real question is whether you need a hosted runtime, a programmable SDK, or an end-user coding surface.
Review comparisonsJudge fit before you commit
Review the use-case and examples pages when you want to sanity-check whether your workload belongs in Managed Agents at all.
See good fitsPlan around operational limits
Read the pricing and limits page before you promise runtimes, network access, or throughput to other teams.
Check limitsUnstick an implementation
Use the troubleshooting page when sessions stall, the beta header is missing, tools do not fire, or the wrong environment assumptions creep in.
Debug the blockerAudience
This site is for teams deciding whether a hosted agent runtime belongs in the stack.
If that is not your decision, one of the comparison pages will tell you what product you probably need instead.
Application engineers
You want a hosted agent runtime without owning the harness, sandbox lifecycle, and event persistence yourself.
Platform and security reviewers
You need a concrete explanation of what runs inside Anthropic-managed infrastructure before you approve a prototype.
Technical leads
You are deciding whether this should be a Managed Agents build, an Agent SDK app, or a Claude Code workflow for your internal team.
Developer advocates and docs owners
You need a user-facing guide that turns scattered docs into a compact path your team can actually follow.
Before You Decide
Check these assumptions before you commit the first sprint.
These questions prevent a lot of architecture churn later.
Treat Managed Agents as a hosted runtime with a built-in agent loop, not as a thin wrapper around the Messages API.
Decide whether you need Anthropic-managed containers or whether your team must keep execution inside your own infrastructure boundary.
Confirm whether the built-in toolset and environment config are enough before you reach for a fully custom SDK stack.
Plan around beta status, org-level spend controls, and per-endpoint rate limits before you promise throughput.
Comparison
Managed Agents only makes sense when you compare it to the real alternatives.
The most expensive mistake is comparing it to a fuzzy idea of “an agent” instead of to the actual Anthropic options.
| Option | What it is | Who owns the runtime | Best for |
|---|---|---|---|
| Messages API | Direct prompts and tool loops you implement yourself. | Your app owns the loop, tools, retries, and persistence. | Teams that want maximum control and can afford the plumbing. |
| Managed Agents | Anthropic-managed agent runtime with sessions, environments, tools, and streaming events. | Anthropic runs the harness and cloud container; you configure the agent and steer sessions. | Long-running, asynchronous tasks when you want hosted infrastructure. |
| Agent SDK | Programmable access to Claude Code tooling and agent loop from CLI, TypeScript, or Python. | You own deployment and environment decisions while reusing Claude Code capabilities. | Custom agents, CI workflows, and agent apps that need code-level control. |
| Claude Code | End-user product for coding in the terminal, IDE, desktop app, and browser. | Anthropic ships the product surface; you use it as an operator, not as infrastructure. | Human-in-the-loop coding work, reviews, and development automation. |
FAQ
These are the quickest answers to the first high-value questions.
Use them to orient yourself before you dive into the longer pages.
Is Managed Agents just Claude Code in the cloud?+
No. Managed Agents is a hosted agent runtime in the Claude Platform. Claude Code is a product surface for humans working in a terminal, IDE, browser, or desktop app.
When is Managed Agents the better fit than building on the Messages API?+
Usually when you need long-running work, built-in event persistence, Anthropic-managed infrastructure, and a smaller amount of orchestration code in your app.
Should I think about limits before the first prototype?+
Yes. Anthropic documents org-level spend limits, endpoint rate limits, and beta requirements. You should treat those as first-class design constraints from day one.