Anthropic details Claude Managed Agents

Anthropic has published developer documentation laying out how “Claude Managed Agents” works, expanding public detail around a product the company launched on April 8 as a hosted service for autonomous, long-running AI tasks. Anthropic’s docs describe the offering as a “pre-built, configurable agent harness” that runs in managed infrastructure and is “best for long-running tasks and asynchronous work.” The May 2026 documentation places Managed Agents alongside Anthropic’s Messages API as one of two main ways to build with Claude. The docs say the Messages API is for developers who want direct model prompting and fine-grained control, while Managed Agents is aimed at users who do not want to build their own agent loop, runtime and tool execution layer. Anthropic said in its April 8 product announcement that Managed Agents is in public beta on the Claude Platform. (platform.claude.com) The company described it there as “a suite of composable APIs for building and deploying cloud-hosted agents at scale.” ### What, exactly, is Anthropic putting in front of developers? Anthropic’s overview page says Managed Agents provides “the harness and infrastructure for running Claude as an autonomous agent.” The docs say developers can configure an agent once, pair it with an environment, launch a session and then send events while Claude executes tools and streams back results. (platform.claude.com) The platform documentation breaks the system into four core concepts: agent, environment, session and events. (claude.com) Anthropic defines the agent as the model, system prompt, tools, MCP servers and skills; the environment as a configured container template; the session as a running agent instance; and events as the messages and status updates exchanged between the application and the agent. ### Which jobs is Managed Agents supposed to handle? (platform.claude.com) Anthropic’s docs say the product is best suited to workloads that run for “minutes or hours” and require multiple tool calls. The same page lists cloud infrastructure, minimal infrastructure burden and stateful sessions with persistent filesystems and conversation history as target use cases. Anthropic’s launch post said the service includes “long-running sessions” that can operate autonomously for hours, with progress and outputs persisting through disconnections. (platform.claude.com) The company also said Managed Agents handles secure sandboxing, authentication and tool execution on its own infrastructure. ### What tools and controls do the docs show? Anthropic’s quickstart shows developers creating an agent, creating an environment and then starting a session through the beta API. (platform.claude.com) The quickstart says Managed Agents requests require a `managed-agents-2026-04-01` beta header, though Anthropic’s SDK sets that header automatically. The same quickstart says the `agent_toolset_20260401` tool type enables a pre-built set of agent tools including bash, file operations and web search. (claude.com) Anthropic’s overview page separately says Claude can read files, run commands, browse the web and execute code securely inside the managed environment. ### How is this different from Anthropic’s regular API? Anthropic’s documentation homepage says Messages gives developers direct model access and requires them to manage conversation state and write their own tool loop. (platform.claude.com) The Managed Agents section says Anthropic instead provides “fully managed agent infrastructure” with stateful sessions and persistent event history. Anthropic’s engineering blog said the company built Managed Agents as a hosted service for “long-horizon agents” through interfaces intended to stay stable even as the underlying harness changes. (platform.claude.com) The post said the system virtualizes three components — session, harness and sandbox. ### What has Anthropic said about rollout and next steps? Anthropic’s April 8 launch post said some features remain limited. The company said multi-agent coordination and outcome-based self-evaluation are available in research preview and require developers to request access. (platform.claude.com) Anthropic’s documentation pages for Managed Agents — including the overview, quickstart and tools references — are live now on the Claude Platform. (anthropic.com) The docs also say the service is available on Claude Platform on AWS, with some differences in feature availability and session behavior. (platform.claude.com) (claude.com)