Public Beta: April 8, 2026

Claude Managed Agents Complete Guide

Anthropic's fully managed AI agent platform — from prototype to production in days, no infrastructure to build

#ClaudeAgents #ManagedAgents #AI Agent #Anthropic

What Are Claude Managed Agents?

Compress months of AI agent infrastructure work down to days

Traditional Agent Development Pain Points

✗Spending dev cycles building secure sandbox infrastructure
✗Manually implementing state management and session persistence
✗Hand-rolling credential security and permission isolation
✗Rewriting the agent loop every time the model is upgraded

Managed Agents Solution

✓Ready-to-use sandboxed code execution environment
✓Automatic checkpointing and session recovery
✓Built-in credential management and scoped permission control
✓Decoupled from model upgrades, automatically adopts new capabilities

Six Core Capabilities

Sandboxed Code Execution

Isolated execution environment — code run by agents executes in a secure sandbox, preventing malicious code from affecting the host system

Checkpointing

Automatically saves agent execution state so long-running tasks can be paused and resumed at any point without restarting

Credential Management

Securely stores and injects API keys, OAuth tokens, and other sensitive credentials that agents access on demand without exposing plaintext

Scoped Permissions

Fine-grained permission control — configure a minimal permission set per agent, following the principle of least privilege

End-to-End Tracing

Complete execution chain tracing — every agent action is observable, auditable, and replayable

Auto-Scaling

Cloud-hosted infrastructure automatically handles concurrency and scaling — no infrastructure ops required

Quick Start

Build your first Managed Agent with the Claude Agent SDK

Python Python SDK

# pip install anthropic

from

anthropic

import

Anthropic

client = Anthropic()

# Create a Managed Agent

agent = client.beta.agents.create(

name="my-agent",

model="claude-opus-4-7",

tools=[{

"type": "computer_20250124"

}]

)

# Create a session and start execution

session = client.beta.agents.sessions.create(

agent_id=agent.id

)

TypeScript TypeScript SDK

// npm install @anthropic-ai/sdk

import

Anthropic

from

'@anthropic-ai/sdk'

;

const client = new Anthropic();

// Create a Managed Agent

const

agent =

await

client.beta.agents.create({

name: 'my-agent',

model: 'claude-opus-4-7',

tools: [{

type: 'computer_20250124'

}]

});

// Create a session and start execution

const

session =

await

client.beta.agents.sessions.create({

agent_id: agent.id

});

Enterprise Adoption

Leading tech companies are already using Managed Agents in production

Notion

Using Managed Agents to build an intelligent knowledge base auto-organization agent, significantly reducing manual documentation work

Asana

Built a project task decomposition and progress tracking agent with Managed Agents to improve team collaboration efficiency

Sentry

Leveraging Managed Agents for automated error log analysis and root cause identification, accelerating bug resolution

Rakuten

Deployed multilingual customer service agents via Managed Agents with automatic Japanese/English/Chinese switching and complex ticket handling

Using Managed Agents via QCode.cc

QCode.cc fully proxies the Anthropic API, including the latest Managed Agents beta endpoints. Developers in China can access all Managed Agents features through QCode.cc without a VPN.

# China users: set QCode.cc relay

export ANTHROPIC_BASE_URL=https://api.qcode.cc

export ANTHROPIC_API_KEY=sk-ant-your-qcode-key

# Call Managed Agents API normally

client = Anthropic() # auto-uses ANTHROPIC_BASE_URL

agent = client.beta.agents.create(...)

Related Resources

Claude Agent SDK →

Developer guide to building custom AI agents with the Agent SDK

多 Agent 编程指南 →

Learn multi-agent programming patterns and Agent Teams collaboration

Claude Code 完全指南 →

Complete guide to the Claude Code CLI tool with advanced tips

Start Building Your Managed Agent

Via QCode.cc relay, developers in China can use all Claude Managed Agents features without barriers