참조

참조 (Reference)

이 페이지는 Claude Managed Agents의 참조 자료를 모아 놓았어요. 이벤트 유형, 셀프 호스팅 워커 CLI 플래그, 지원되는 MCP 서버 유형, 속도 제한, 그리고 브랜딩 가이드라인까지 한눈에 볼 수 있어요. 작업 지향 가이드가 필요하면 각 섹션의 링크를 따라가세요. 세션 리소스의 작업(operation)은 Session operations에서 다루어요.

출처: 문서

본문

이 페이지는 Claude Managed Agents의 참조 자료를 모아 놓아요. 작업 지향 가이드는 각 섹션의 링크를 따라가세요. 세션 리소스의 작업은 Session operations을 참고하세요.

Event types

영구 저장되는 이벤트 유형 문자열은 {domain}.{action} 명명 규칙을 따라요. 스트림 전용 이벤트 델타(Event deltas 탭 참고)는 예외예요. 보내기·스트리밍·나열은 Session event stream을, 웹훅 이벤트 유형은 Subscribe to webhooks에서 별도로 다루며, 일부 이름이 스트림과 다를 수 있어요(예: session.status_idled vs session.status_idle).

| Type | Description | | ------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `user.message` | A user message with text, image, or document content. | | `user.interrupt` | Stop the agent mid-execution. | | `user.custom_tool_result` | Response to a custom tool call from the agent. | | `user.tool_confirmation` | Approve or deny an agent or MCP tool call when a permission policy requires confirmation. | | `user.define_outcome` | Define an [outcome](https://platform.claude.com/docs/en/managed-agents/define-outcomes) for the agent to work toward. | | `user.tool_result` | For sessions with `self_hosted` [environments](https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes) only, your integration is responsible for providing `agent_toolset` results. The SDK helpers and CLI do this automatically. | | Type | Description | | -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `agent.message` | Agent response content blocks. | | `agent.thinking` | Signals the agent is making forward progress through extended thinking. This is a progress signal only and does not carry the thinking content. | | `agent.tool_use` | Agent invokes a pre-built agent tool (bash, file operations, and so on). Carries `evaluated_permission` and, usually, `evaluation` (see [how each call was evaluated](https://platform.claude.com/docs/en/managed-agents/permission-policies#see-how-each-call-was-evaluated)). | | `agent.tool_result` | Result of a pre-built agent tool execution. | | `agent.mcp_tool_use` | Agent invokes an MCP server tool. Carries `evaluated_permission` and, usually, `evaluation` (see [how each call was evaluated](https://platform.claude.com/docs/en/managed-agents/permission-policies#see-how-each-call-was-evaluated)). | | `agent.mcp_tool_result` | Result of an MCP tool execution. | | `agent.custom_tool_use` | Agent invokes one of your custom tools. Respond with a `user.custom_tool_result` event. | | `agent.thread_context_compacted` | Conversation history was compacted to fit the context window. | | `agent.thread_message_received` | In a [multiagent](https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration) session, a message from another thread arrived on the thread whose stream carries this event; on the primary thread, an agent sent a report or question to the coordinator. | | `agent.thread_message_sent` | In a [multiagent](https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration) session, the thread whose stream carries this event sent a message to another thread; on the primary thread, the coordinator sent a task or follow-up message to another agent. | 
Message content in these events can include a `redacted` content block, `{"type": "redacted"}`: a placeholder for content withheld by Anthropic model policy. The block carries no other fields. Redacted blocks appear only in content the platform emits; a user event that includes one is rejected with a 400 error.
| Type | Description | | ----------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `session.status_running` | Agent is actively processing. | | `session.status_idle` | Agent finished its current task and is waiting for input. Includes a `stop_reason` indicating why the agent stopped. | | `session.status_rescheduled` | A transient error occurred and the session is retrying automatically. | | `session.status_terminated` | Session ended, either because of an unrecoverable error or because it was archived. | | `session.deleted` | Session was deleted. Terminates any active event stream; no further events are emitted for this session. | | `session.updated` | Session update request changed at least one field. Includes only the fields that changed. Updates apply on the next turn. | | `session.error` | An error occurred during processing. Includes a typed `error` object with a `retry_status`. | | `session.usage` | Snapshot of the session's cumulative usage and tracked list cost. Carries the session's usage totals and an echo of the session's [budget](https://platform.claude.com/docs/en/managed-agents/budgets), or `null` when the session has none. | | `session.thread_created` | A [multiagent](https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration) thread was created. | | `session.thread_status_running` | A session thread began executing. Every session emits this for its primary thread; in [multiagent](https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration) sessions, child-thread transitions are also cross-posted to the primary stream. | | `session.thread_status_idle` | A session thread finished its turn and is awaiting input. Includes `stop_reason`. | | `session.thread_status_rescheduled` | A session thread hit a transient error and is retrying automatically. | | `session.thread_status_terminated` | A session thread was archived or reached a terminal error. | Span events are observability markers that wrap activity for timing and usage tracking. 
| Type                              | Description                                                                                                                                                                                                                                              |
| --------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `span.model_request_start`        | A model inference call has started.                                                                                                                                                                                                                      |
| `span.model_request_end`          | A model inference call has completed. Includes `model_usage` with token counts.                                                                                                                                                                          |
| `span.outcome_evaluation_start`   | [Outcome](https://platform.claude.com/docs/en/managed-agents/define-outcomes) evaluation has started.                                                                                                                                                    |
| `span.outcome_evaluation_ongoing` | Heartbeat during an ongoing [outcome](https://platform.claude.com/docs/en/managed-agents/define-outcomes) evaluation.                                                                                                                                    |
| `span.outcome_evaluation_end`     | An [outcome](https://platform.claude.com/docs/en/managed-agents/define-outcomes) evaluation cycle has completed. A `needs_revision` result means another cycle follows; `satisfied`, `max_iterations_reached`, `failed`, and `interrupted` are terminal. |
| Type | Description | | ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `system.message` | Append privileged system-level context that applies to the accompanying turn and all subsequent turns. Supported on Claude Fable 5.1, Claude Mythos 5.1, Claude Fable 5, Claude Mythos 5, Claude Opus 5.5, Claude Opus 5, and Claude Opus 4.8. On an unsupported primary model the event is rejected with `model_does_not_support_mid_conversation_system`. | Event deltas are stream-only preview events. They are emitted on stream connections (session-level or per-thread) that opt in with the `event_deltas[]` parameter, and they are never persisted to the session's event history. See [Event deltas](https://platform.claude.com/docs/en/managed-agents/events-and-streaming#event-deltas) for opting in, accumulating, and reconciling them. 
| Type          | Description                                                                                                              |
| ------------- | ------------------------------------------------------------------------------------------------------------------------ |
| `event_start` | A previewed event has started generating. Carries the upcoming event's `type` and `id`. Stream-only and never persisted. |
| `event_delta` | Incremental content for a previewed event, identified by `event_id`. Stream-only and never persisted.                    |

Self-hosted worker

self_hosted 환경을 구동하는 미리 빌드된 워커를 위한 ant beta:worker CLI 플래그들이에요. 환경 설정, 워커 실행, SDK 헬퍼 옵션은 Self-hosted sandboxes를 참고하세요.

Flag Description
--environment-id The environment to poll for work. Also reads from ANTHROPIC_ENVIRONMENT_ID.
--environment-key Authenticates the worker with this environment. Also reads from ANTHROPIC_ENVIRONMENT_KEY.
--workdir Directory where skills are downloaded and tools read and write files. Defaults to . (the current directory); the system default working directory is /workspace.
--on-work Script to call for each claimed work item instead of running tools in-process. Receives session details as environment variables.
--unrestricted-paths Allow the file tools to read and write paths outside --workdir. The workdir check is a guardrail for the file tools only, not a sandbox; it does not constrain bash.
--max-idle How long to wait after the session goes idle with an end_turn stop reason before shutting down. Defaults to 60s.
--log-format Log output format. Use json for structured log ingestion. Defaults to text.

CLI 워커는 메모리 스토어를 마운트하지 않아요: 하나를 붙인 세션은 여전히 실행되지만, 에이전트는 스토어의 mount_path에서 아무것도 찾지 못하고 변경도 스토어로 다시 동기화되지 않아요. 셀프 호스팅 환경의 세션에서 메모리 스토어를 사용하려면 대신 SDK 워커를 실행하세요. Use memory stores를 참고하세요.

Supported MCP server types

Claude Managed Agents는 HTTP 엔드포인트를 노출하는 원격 MCP 서버에 연결하거나, MCP tunnels을 통해 프라이빗 MCP 서버에 연결해요. 서버는 MCP 프로토콜의 streamable HTTP 전송을 지원해야 해요. deprecated된 SSE 전송만 지원하는 서버는 자동 폴백을 통해 여전히 작동해요. 에이전트에 서버를 선언하는 방법은 MCP connector를 참고하세요.

MCP와 MCP 서버 구축에 대한 자세한 내용은 MCP documentation을 참고하세요.

Rate limits

Managed Agents 엔드포인트는 조직별로 속도가 제한돼요:

Operation Limit
Create endpoints (such as agents, sessions, and environments) 300 requests per minute
Read endpoints (such as retrieve, list, and stream) 1,200 requests per minute

조직 수준의 spend limits and usage-tier rate limits도 적용돼요.

Branding guidelines

Claude Managed Agents를 통합하는 파트너는 Claude 브랜딩 사용이 선택 사항이에요. 제품에서 Claude를 언급할 때:

Allowed:

  • "Claude Agent" (preferred for dropdown menus)
  • "Claude" (when within a menu already labeled "Agents")
  • "{YourAgentName} Powered by Claude" (if you have an existing agent name)

Not permitted:

  • "Claude Code" or "Claude Code Agent"
  • "Claude Cowork" or "Claude Cowork Agent"
  • Claude Code-branded ASCII art or visual elements that mimic Claude Code

제품은 자체 브랜딩을 유지하고 Claude Code, Claude Cowork 또는 다른 Anthropic 제품처럼 보이면 안 돼요. 브랜딩 준수에 관한 질문은 Anthropic sales team에 연락하세요.

더 알아보기 (Learn more)