Anthropic publishes Claude Managed Agents docs

Anthropic has published developer documentation for Claude Managed Agents, a new documented surface in the Claude API that the company describes as a pre-built, configurable agent harness for long-running tasks and asynchronous work. The documentation appeared on Anthropic’s Claude API site on May 14 and lays out how developers can define agents, configure environments and run stateful sessions in Anthropic-managed infrastructure. Anthropic says the system is designed to handle tool execution and runtime management without requiring developers to build their own agent loop. The docs also describe memory, session streaming and configuration options for developers using the API. ### What exactly did Anthropic publish? Anthropic’s Claude API documentation now includes a dedicated Managed Agents section with an overview, quickstart and API reference pages. The overview page contrasts the new surface with the Messages API, saying Messages is for direct model prompting and custom agent loops, while Managed Agents is for long-running tasks and asynchronous work. The quickstart page says developers can create an agent, set up an environment, start a session and stream responses. (platform.claude.com) Anthropic also points developers to Console-based onboarding, where agents can be created and tested visually before code is written. ### How does Anthropic describe the product? Anthropic says Claude Managed Agents “provides the harness and infrastructure for running Claude as an autonomous agent.” The company says developers get a managed environment in which Claude can read files, run commands, browse the web and execute code securely, rather than having to assemble those pieces themselves. (platform.claude.com) The docs say the harness includes built-in prompt caching, compaction and other performance optimizations. (platform.claude.com) Anthropic’s API overview lists Agents, Sessions and Environments as separate beta APIs tied to the Managed Agents surface. ### What are the building blocks developers have to configure? Anthropic’s overview page says Managed Agents is built around four concepts: agent, environment, session and events. (platform.claude.com) In the documentation, an agent includes the model, system prompt, tools, MCP servers and skills; an environment is a configured container template; a session is a running agent instance; and events are the messages and updates exchanged between the application and the agent. The API overview page says the Agents API is used to define reusable, versioned agent configurations, the Sessions API runs stateful agent sessions in managed cloud containers, and the Environments API configures container templates for those sessions. That places the feature inside Anthropic’s broader REST API rather than as a separate product line. ### Where do memory and persistence fit in? Anthropic’s memory documentation says each Managed Agents session starts with a fresh context by default, and information learned during a session is lost when that session ends unless a memory store is attached. (platform.claude.com) The company says memory stores are workspace-scoped collections of text documents that can persist information such as user preferences, project conventions and prior mistakes across sessions. (platform.claude.com) A separate memory-tool page in Anthropic’s broader tools documentation describes client-side memory for Claude across conversations, but the Managed Agents memory page describes a hosted pattern in which a store is mounted into the session container. The distinction in the docs is that Managed Agents memory is part of the managed session workflow Anthropic is documenting here. ### What does the documentation say about availability and status? (platform.claude.com) Anthropic’s quickstart and memory pages say all Managed Agents API requests require the `managed-agents-2026-04-01` beta header, and the API overview lists the related endpoints under a beta section. The company’s Claude API skill documentation also describes Managed Agents as beta and says the feature is available on the Claude API and Claude Platform on AWS, with differences in feature availability and session behavior noted in the docs. (platform.claude.com) Anthropic’s next step for developers is laid out in the docs themselves: the quickstart walks through first-session setup, while the sessions, environments and memory pages provide the implementation details for running agents in managed containers and persisting data across runs. (platform.claude.com)