HeadsUpAI

Cloudflare Integrates Claude Managed Agents to Secure Autonomous Code Execution

Cloudflare integrated with Anthropic to host Claude Managed Agents within its isolated execution environments. This setup decouples the reasoning loop on Anthropic's platform from the code execution on Cloudflare. Developers can now deploy Claude Managed Agents into Cloudflare Sandboxes to run untrusted code securely.
Availability
Available now via Cloudflare dashboard
Sandbox types
MicroVMs (Containers) and V8 Isolates (Dynamic Workers)
Startup time (Isolates)
Milliseconds
Connectivity
Cloudflare Mesh, Workers VPC, and outbound proxies
Built-in tools
Browser Run, Agent Inboxes, and Workers AI

This integration addresses the security and scaling bottlenecks of autonomous systems, mirroring Anthropic's self-hosted sandboxes for private infrastructure. By using Cloudflare's isolate-based runtime, agents can boot in milliseconds and scale to millions of concurrent sessions. It also enables zero-trust connectivity, allowing agents to access private internal services without exposing them to the internet.

You can use the new deployment template to give agents dedicated email addresses, programmable browsers, and secure credential injection through outbound proxies. The system supports two backend types: full MicroVMs for complex engineering tasks and lightweight V8 isolates for high-speed workflows. The integration is available now through the Cloudflare dashboard and GitHub.

View the full update on blog.cloudflare.com
Cloudflare
Cloudflare
@Cloudflare
X

Cloudflare has integrated with Anthropic's Claude Managed Agents to provide a fast, isolated execution environment for autonomous code delivery. This means builders can scale agent workflows globally while strictly controlling access to private backends and easily customizing their agent’s tools and runtimes. https://t.co/HZ1kD0qVmO

60retweets378likes
View on X

Still wondering? A few quick answers below.

This integration allows developers to run the core reasoning loop of an AI agent on Anthropic's platform while using Cloudflare's infrastructure for code execution and tool use. By decoupling the agent's brain from its hands, builders can maintain strict control over security, private data connectivity, and the specific runtime environment where the agent operates.

When a Claude Agent starts a session, it connects to a Cloudflare-based control plane that spins up a secure sandbox. Developers can choose between full MicroVMs for complex engineering tasks or lightweight V8 isolates for high-scale workflows. These environments automatically persist state across sessions and can be observed or accessed via SSH through the Cloudflare dashboard.

Agents can securely access internal data and services without exposing them to the public internet by using Cloudflare Mesh or Workers VPC. This setup uses customizable outbound proxies to perform zero-trust authentication and inject encrypted credentials into requests. This architecture prevents sensitive secrets from being exposed directly to the agent, protecting against potential data exfiltration attacks.

The integration includes native support for Browser Run, which provides agents with programmable browsers for navigating the web and recording sessions. Agents can also be assigned dedicated email addresses to send and receive messages. Other available tools include image generation via Workers AI and the ability to create custom tools using standard JavaScript functions.

The integration is available now for developers using both platforms. Cloudflare provides a default deployment template and an onboarding guide on GitHub to help users set up the environment in minutes. While the core integration is accessible, specific features like Dynamic Workers or specialized container images may depend on your existing Cloudflare service tier.

Share this update