Cloudflare and Anthropic add Claude agents to Cloudflare Sandboxes
Cloudflare and Anthropic have integrated Claude Managed Agents with Cloudflare Sandboxes, allowing developers to run agent execution environments on Cloudflare's platform while the core agent loop remains on Anthropic. This integration provides enhanced security, sandbox control, and scalability options, including lightweight V8 isolates for high-volume agent workloads. It also enables agents to interact with web browsers via Browser Run and send emails with built-in tools.
Key Takeaways
- The integration keeps the Claude agent loop on Anthropic while Cloudflare handles execution, security, and custom tool calls.
- Cloudflare says lightweight V8 isolate sandboxes can boot in milliseconds and handle tens of thousands of concurrent agents.
- The default deployment template includes Browser Run tools such as browser_search, browser_execute, screenshot, browse, fetch_to_markdown, and web_fetch.
- Built-in email tools include send_email, email_read, and email_list, plus Cloudflare Email Service support for agent inboxes.
- Private-service access can use Cloudflare Mesh or Workers VPC with outbound proxies, header injection, and egress policies.
Why It Matters
This gives developers a split architecture for Claude agents: Anthropic runs the core loop, while Cloudflare supplies sandboxes, browser automation, email, and private connectivity. That matters for teams that want more control over security, compliance, and observability without rebuilding the agent stack. It also broadens Cloudflare’s developer platform around agents, pairing Sandboxes, Agents SDK, Browser Run, Dynamic Workers, Containers, Mesh, and Workers VPC. The next signal to watch is how quickly teams choose the “isolate” backend versus microVMs when deploying at scale.
Read full article at blog.cloudflare.com