Claude Managed Agents is a hosted agent harness: Anthropic runs the agent loop, the sandbox, and tool execution, and your application sends events into a running session instead of implementing its own runtime.1 It was in beta at review time (2026-09-15); every request needs the managed-agents-2026-04-01 header, which the SDK sets.1

Building blocks

ConceptWhat it is
AgentModel, system prompt, tools, MCP servers, and skills; created once and referenced by ID
EnvironmentWhere sessions run: an Anthropic-managed cloud sandbox or a self-hosted one
SessionA running agent instance in an environment, doing one task
EventsMessages between your application and the agent: user turns, tool results, status updates

As described in the documentation.1 You create an agent and an environment, start a session, and send events; Claude runs tools on its own and streams results back over server-sent events. Event history is stored server-side, so a session survives pauses, and you can steer or interrupt it with further events.1

Built-in tools are Bash, file operations (read, write, edit, glob, grep), web search and fetch with optional domain allow or block lists, and MCP servers.1

When it fits

Use it for long-running work (minutes or hours across many tool calls), managed or self-hosted sandboxes, stateful sessions with a persistent filesystem, and cron-scheduled runs. Stay on the Messages API when you need a custom agent loop or control over every model call.1

Constraints at review time

  • Access was on by default for API accounts; MCP tunnels and “dreaming” were a narrower research preview.1
  • History, sandbox state, and outputs persist server-side, so it was not eligible for Zero Data Retention or HIPAA BAA coverage. Sessions and uploaded files can be deleted through the API.1
  • Behaviour could still change between beta releases.1

Footnotes

  1. Claude Managed Agents 2 3 4 5 6 7 8 9