Announcing Claude Managed Agents on Cloudflare

Cloudflare and Anthropic announced an integration of Claude Managed Agents with Cloudflare Sandboxes. The companies said the integration gives developers more control over agent sandboxes, secures connections to private services, and improves observability. Cloudflare said its Developer Platform has expanded over the past year to provide tools to run agents at scale, including Sandboxes for stateful Linux microVMs, an Agents SDK, Browser Run for programmable browsers, and Dynamic Workers for dynamic sandboxed code execution. The integration allows developers to run the agent loop on the Claude Platform while using Cloudflare to execute code, secure connections, and run custom tool calls. A default deployment template provides enhanced security through customizable proxies, sandbox control and observability with detailed metrics and SSH access, lightweight sandboxes using V8 isolates or microVMs, private service connectivity without Internet exposure, browser session recording, email capabilities, and custom tools. Anthropic describes the approach as decoupling the brain from the hands, with the core agent loop running on Anthropic infrastructure and code execution infrastructure running on Cloudflare. The control plane uses Workers to give each agent session a sandboxed environment with automatically persisted state. Sandbox logs can be queried or shipped to external providers. Cloudflare Mesh and Workers VPC enable connections to private services using post-quantum encrypted networking without a VPN or bastion host.