Historical source material recovered on 2026-09-09. Capture dates and source limitations apply; recovery is not fresh verification.

Claude Managed Agents

Anthropic-hosted agent platform launched April 2026 that lets users define an agent (name, system prompt, model, tools, MCP connections, skills) and deploy it to an Anthropic-managed cloud sandbox. Targets non-engineers who want production agents without DevOps. Includes a credential vault with OAuth-based MCP connections, environment isolation, and a CLI interface for building managed agents from inside Claude Code.

Key facts

• Billing: 8 cents/hour of active session time plus API token costs
• Credential vault stores OAuth MCP connections shareable across teams
• Each agent runs inside a cloud container environment with networking rules
• Supports model, system prompt, MCP server, and skills configuration
• No native cron triggers at launch — agents must be invoked by API call
• Three teased features require early-access application: outcomes (self-eval), multi-agent orchestration, persistent memory
• Notion shipped a public integration where dragging tickets triggers managed agents
• Nate's later deployment comparison frames managed agents as a good Anthropic-hosted path for users who do not want to operate their own runtime, while noting that experienced Claude Code users may prefer direct control over repo, infra, tools, and deployment surfaces ^[src-086].
• Anthropic's June 2026 article reframes Managed Agents as production infrastructure: hosting, scaling, sessions, filesystem state, execution isolation, credentials, and observability are treated as core agent requirements, not optional DevOps details ^[src-114].
• The article says Managed Agents separates the harness that calls Claude from the sandbox where code executes, with sessions acting as append-only logs of model calls, tool calls, and results ^[src-114].
• Anthropic describes three main resources: agents, environments, and sessions. It also highlights Vaults for credentials, persistent sessions, Memory and Dreaming, self-hosted sandboxes, MCP tunnels, outcomes, multi-agent orchestration, permission policies, and webhooks ^[src-114].
• Anthropic reports that this architecture reduced time-to-first-token by roughly 60% at p50 and more than 90% at p95 in its testing ^[src-114].

Source references

• ^[src-004] Nate Herk cluster — Nate Herk — Claude Code cluster (21 videos)

– Videos referenced: 27Y44JYXZJ8

• ^[src-086] Nate Herk — "I Tested 3 Ways to Deploy Claude Agents (Here's When to Use Each)" (2026-05-15)
• ^[src-114] Anthropic – "The evolution of agentic surfaces: building with Claude Managed Agents" (2026-06-10)

Historical recovered snapshot — 2026-08-03

Preserved from pr-123. Documentary recovery on 2026-09-09; claims below retain their historical source scope and were not freshly verified. Current synthesis above remains authoritative.

• ^[src-012] Nate Herk cluster — Nate Herk — Claude Code cluster (21 videos)

– Videos referenced: 27Y44JYXZJ8

• ^[src-086] Nate Herk — "I Tested 3 Ways to Deploy Claude Agents (Here's When to Use Each)" (2026-05-15)
• ^[src-114] Anthropic – "The evolution of agentic surfaces: building with Claude Managed Agents" (2026-06-10)

• The managed-agent preview added Gemini 3.6 Flash, environment hooks, token budgets, scheduled triggers and free-tier access ^[src-277].