Cloudflare has added support for Claude Managed Agents. What this means: you can keep the "brain" of your Claude agent on Anthropic's platform while running the "hands" (code execution, data access, and so on) on Cloudflare's infrastructure. For teams that need to securely connect to private systems and meet compliance requirements, this is a practical integration path.

What This Is

Anthropic describes the architecture as "decoupling the brain from the hands":

• The brain (agent logic and orchestration): stays on Anthropic's platform
• The hands (code execution layer): runs on Cloudflare's infrastructure

Previously, using Claude Managed Agents meant running the entire stack on infrastructure provided by Anthropic. Now you can choose to move the execution layer to Cloudflare and gain greater control over your infrastructure.

Core Capabilities

Sandboxed Execution Environments

• Lightweight execution: fast-start sandboxes built on Cloudflare Workers
• Virtual machine execution: for scenarios demanding stronger isolation, VM-level sandboxes are available
• Both modes ship with sandbox controls and logging built in

Private Service Access

Through Cloudflare Mesh and Workers VPC services, agents can securely connect to your private systems without exposing services to the public internet:

• Private connections never touch the public internet
• Credentials are injected; the agent itself never holds them
• Agent operations are logged for auditability

Built-In Capabilities

• Browser automation: the Browser Run feature handles web operations
• Email functionality: built-in mail send and receive
• Custom tools: define the toolset your agents can call
• Logging and monitoring: full activity logs

Real-World Scenarios

Financial Services

Cloudflare regional director Sebastian Weiss offered a concrete scenario:

The agent runs inside the bank's own cloud environment and accesses core systems over a private connection that never touches the public internet. Credentials are injected, and agent operations are logged for auditors to review. Claude holds the decision-making power; control always stays with the bank.

Internal Enterprise System Integration

Teams that need AI agents to reach internal CRMs, ERPs, or databases can use Workers VPC to establish secure channels, with agents connecting only to approved internal services.

Custom AI Agent Workflows

Extend agent capabilities with Cloudflare's developer platform:

• Artifacts: Git-based code repository management
• Workers AI: edge inference
• Dynamic Workers: on-demand application hosting

How to Get Started

Prerequisites

• A Cloudflare account
• An Anthropic API key (Claude access)
• Basic Workers development experience

Quick Start

Cloudflare provides an open-source deployment template as a quick start. Developers can customize the control plane on top of the template, then configure network proxies, secure connections, and toolsets according to their business needs.

On costs, developers manage expenses related to Claude API usage and Cloudflare resources separately.

How It Compares to Other Platforms

Cloudflare isn't the only provider supporting self-hosted Claude Managed Agents. The launch lineup also includes:

• Daytona: development environment management
• Modal: serverless compute
• Vercel: frontend deployment

Among them, Cloudflare occupies the "prime position," offering both Browser Run automation and quick-start templates.

Who It's For

• Enterprises with compliance requirements: industries like finance and healthcare with strict data-security demands
• Teams connecting to private systems: anyone who doesn't want internal services exposed to the public internet
• Developers who want infrastructure control: those who, for performance, cost, or security reasons, want to choose their own execution environment