Claude Managed Agents on Cloudflare
13 hours ago
- #Anthropic
- #Cloudflare
- #AI Agents
- Cloudflare and Anthropic integrated Claude Managed Agents with Cloudflare Sandboxes for enhanced control, security, and observability.
- Features include enhanced security via customizable proxies, sandbox control with metrics/logs, and lightweight sandboxes (microVMs or isolates) for scaling.
- Provides private service connectivity, browser control with session recording, email capabilities, and custom tools without extra infrastructure.
- Claude Managed Agents decouple the 'brain' (agent loop on Anthropic) from the 'hands' (execution on Cloudflare), offering flexibility.
- Cloudflare environment includes a Workers-based control plane for sandboxed code execution, with state persistence and customization options.
- Lightweight V8 isolate sandboxes enable faster boot times and massive scale, compared to microVM-based sandboxes for full applications.
- Security features include outbound proxies for zero-trust authentication, Cloudflare Mesh/VPC for private services, and egress policies.
- Built-in tools include Browser Run for web interaction, email tools, and services like image generation via Workers AI.
- Custom tools can be added easily by forking the repository and writing functions, leveraging Cloudflare's Developer Platform.