Cloudflare brings Claude Managed Agents onto its own sandbox stack for tighter control

Claude Managed Agents

Cloudflare and Anthropic have integrated Claude Managed Agents with Cloudflare Sandboxes, giving developers a way to keep Claude’s managed agent loop on Anthropic while running code, browser sessions, private-service access, and custom tools on Cloudflare infrastructure.

# Cloudflare brings Claude Managed Agents onto its own sandbox stack for tighter control

## Opening summary

Cloudflare says it has integrated Claude Managed Agents with Cloudflare Sandboxes, giving developers a way to run Anthropic’s managed agent harness while keeping execution inside Cloudflare’s own environment. The company is pitching the setup as a way to secure connections to private services, improve observability, and customize how autonomous code runs.

## Main article

The announcement matters because it separates the model layer from the runtime layer in a more explicit way. Cloudflare says developers can keep the agent loop on Anthropic’s Claude platform while shifting execution onto Cloudflare, where code, browser sessions, network paths, and supporting tools can be controlled more directly.

Cloudflare says its default template includes customizable proxies, sandbox logs and metrics, SSH access, lightweight sandbox options, private service connectivity, browser-session auditing, per-agent email, and custom tools. In other words, the integration is not just a connector but a packaged execution environment for teams that want more say over where agent actions happen.

Anthropic’s own documentation helps explain why that matters. The company describes Claude Managed Agents as a pre-built, configurable harness for long-running and asynchronous work, and says environments can run either in Anthropic-managed cloud containers or in self-hosted sandboxes on customer-controlled infrastructure. That makes Cloudflare’s pitch less about novelty alone and more about slotting itself in as the execution layer for customers that need compliance, data-residency, or operational control.

Cloudflare also frames the design as giving developers a way to let Claude handle the reasoning while Cloudflare handles the hands-on work. For AI teams building autonomous coding, browsing, and tool-using systems, that split may become a more important product choice than raw model capability on its own.

## Why it matters

Managed agents are turning into platform architecture decisions, not just chatbot features. Cloudflare wants to own the runtime side of that stack, and this release gives Anthropic customers a more concrete way to choose where their agents actually act.

## Source notes

- Cloudflare says the integration connects Claude Managed Agents with Cloudflare Sandboxes and includes controls for proxies, private connectivity, browser observability, custom tools, email, and sandbox metrics. - Anthropic describes Claude Managed Agents as a pre-built, configurable harness for long-running and asynchronous work. - Anthropic says Managed Agents environments can run in Anthropic-managed containers or self-hosted sandboxes on infrastructure customers control.