Claude Managed Agents
Anthropic-managed agent harness — Agent, Environment, Session, Events
Claude Managed Agents is Anthropic’s pre-built, configurable agent harness on managed infrastructure. You define an Agent (model, system prompt, tools, MCP servers, skills), an Environment (Anthropic-managed cloud sandbox or a self-hosted sandbox on your infrastructure), then start a Session that performs a task and streams Events (user turns, tool results, status). Built-in tools: bash, file operations (read, write, edit, glob, grep), web search and fetch (optional domain allowlist/blocklist), and MCP servers. Sessions are stateful: conversation history, sandbox state, and outputs persist server-side; you can steer or interrupt mid-run. This is the managed-runtime path. Complementary: OpenAI Agents SDK (in-app harness you host, not OpenAI’s managed Agents API) and LangChain Managed Deep Agents (LangSmith-hosted Deep Agents plus Connections). Primary: https://docs.anthropic.com/en/docs/managed-agents/overview.
Use Cases
- Long-running or asynchronous agent tasks (minutes to hours, many tool calls)
- Run Claude in an Anthropic cloud sandbox without building your own agent loop
- Self-host the sandbox for compliance or data-residency while using the managed harness
- Steer or interrupt a live session with additional user events
- Connect MCP servers and built-in bash / file / web tools in one agent config
Key Features
Four core concepts
Agent (model, prompt, tools, MCP, skills), Environment (cloud or self-hosted sandbox), Session (running instance), Events (messages exchanged with the agent, persisted server-side).
Built-in toolset
Bash, file operations (read, write, edit, glob, grep), web search and fetch, MCP servers. Docs: agent_toolset_20260401; web tools accept domain allow/block lists.
Beta header managed-agents-2026-04-01
Claude Managed Agents is in beta. All Managed Agents endpoints require the managed-agents-2026-04-01 beta header (the SDK sets it). Behaviors may change between releases. Within the beta, MCP tunnels and dreaming are a more limited research preview.
Not currently ZDR / HIPAA BAA eligible
Because sessions are stateful (history, sandbox state, outputs stored server-side), Anthropic documents that Managed Agents is not currently eligible for Zero Data Retention or HIPAA Business Associate Agreement coverage. You can delete sessions and separately delete uploaded files through the API. See Anthropic’s API and data retention docs for eligibility across features.
Complementary to other catalog runtimes
OpenAI Agents SDK is an in-app loop you host. LangChain Managed Deep Agents is LangSmith’s managed Deep Agents runtime plus Connections. This row is Anthropic’s managed harness. Do not collapse the three.