AI Agent Documentation
Summary
Docs for Anthropic's Managed Agents multi-agent orchestration: a coordinator delegates work to agents in one session, each with its own isolated thread, model, tools, and prompts, sharing only the sandbox and filesystem.
Highlights
- A coordinator agent delegates to other agents, each running in its own persistent, context-isolated session thread
- All agents share the same sandbox, filesystem, and vault credentials, but tools, MCP servers, and context are not shared
- Each agent has its own configuration: model, system prompt, tools, MCP servers, and skills
- Recommended delegation patterns: parallelization, specialization, and escalation to more capable models
- The multiagent.agents roster accepts agent references by ID, pinned versions, or self for spawning copies of the coordinator
- Requires the managed-agents-2026-04-01 beta header, set automatically by the SDK
auto-generated
Context
Audience
Developers building applications with Anthropic's Managed Agents API who need to orchestrate multiple AI agents
DomainAI Agent Orchestration
Formattechnical documentation
Accessfree online
Related
Anthropic Managed Agents APIClaude PlatformMCP serversagent parallelizationmulti-agent patterns