Cloudflare and Anthropic have integrated Claude Managed Agents with Cloudflare Sandboxes, giving developers more control over agent infrastructure, secure connections to private services, and improved observability. The collaboration addresses growing demand for flexible agent deployment options beyond Anthropic-provided infrastructure.
Claude Managed Agents allow developers to define and run agents on the Anthropic platform where the model can read files, run commands, browse the web, and execute code. Until now, the entire stack ran on Anthropic infrastructure. The new integration decouples the agent loop (the “brain”) from code execution infrastructure (the “hands”), enabling deployment across multiple cloud providers including Cloudflare.
Deployment and Control Plane Setup
The integration provides a default deployment template that can be configured in minutes. When a Claude Agent starts a session, it sends a message to a Workers-based control plane that provisions a sandboxed environment for executing code, developing applications, and running CLI tools. State automatically persists across session sleeps.
Developers can optionally configure sandbox instance sizes or customize container images within VM-based sandboxes. Each sandbox appears in the Cloudflare dashboard where logs can be queried or shipped to external providers like Datadog or Splunk. The control plane includes a built-in UI for tracking sandbox state and SSH access to running machines.
Lightweight Isolate Option for Scale
The integration offers two sandbox options: traditional microVMs for agents requiring full Linux environments and custom tools, or lightweight V8 isolates for faster, cheaper deployments. Isolates execute arbitrary code within a file system but boot in milliseconds rather than seconds, reducing infrastructure costs significantly.
Isolate-based sandboxes allow handling tens of thousands of concurrent agents simultaneously, enabling scaling patterns unavailable with VM-only solutions. Developers select “isolate” as the backend type during agent setup to access this lightweight primitive.
Security and Private Service Access
The integration secures agent traffic through customizable outbound proxies that inject credentials into requests outside sandboxes, preventing agent access to secrets and protecting against exfiltration attacks. Agents can connect to private internal services via Cloudflare Mesh or Workers VPC without exposing those services to the internet.
Egress policies can restrict access per tenant, per agent, or based on custom metadata. Developers can allowlist specific endpoints, perform zero-trust credential injection, and write custom proxy middleware for enhanced security. Post-quantum encrypted networking supports secure connections without requiring VPN or bastion hosts.
Built-in Tools and Browser Control
The deployment ships with multiple pre-configured tools including browser_search, browser_execute, screenshot, and browse functions. Browser Run provides agents with fully programmable and observable browsers running on Cloudflare infrastructure, allowing session recording and audit trails of every action.
Additional built-in capabilities include email tools (send_email, email_read, email_list) allowing agents to send and receive messages using Cloudflare Email Service. Agents can kick off new sessions via email or reply to forwarded messages. The integration also supports call_service for private service connections and image_generate, which uses Workers AI for image generation.
Custom Tool Development and Developer Platform Integration
Developers can fork the integration repository to add customized tools without additional infrastructure. Custom tools are defined using simple functions in configuration files. The Cloudflare Developer Platform provides integrations with R2 object storage for file hosting, Artifacts for git-backed repositories, Workers AI for edge inference, and Dynamic Workers for hosting applications.
Cloudflare’s Developer Platform has expanded over the past year to include Sandboxes for stateful Linux microVMs, an Agents SDK for customizable agent frameworks, Browser Run for programmable browsers, and Dynamic Workers for sandboxed code execution. The integration with Claude Managed Agents extends these capabilities across the artificial intelligence development landscape.
Availability and Next Steps
The Claude Managed Agents integration on Cloudflare is available immediately. Developers can follow the onboarding guide to deploy a working environment, then fork and customize the repository as needed. The integration supports deployment at internet scale, whether running tens of millions of agents using isolates or implementing complex security configurations for enterprise use cases.
