Cloudflare and Anthropic launch 'Environments for Claude' to run Claude-powered apps at the edge

Cloudflare said on May 19 that it is collaborating with Anthropic to launch Cloudflare Environments for Claude Managed Agents, a product that lets companies run Claude agent workloads inside Cloudflare-managed execution environments rather than only on Anthropic’s own infrastructure. The companies said the setup combines Anthropic’s managed agent loop with Cloudflare Workers, private networking and sandboxing controls for code execution. Anthropic announced the broader update on the same day as part of new security features for Claude Managed Agents. The release targets a problem enterprise buyers have raised around agent deployments: how to let an AI system call tools and internal services without exposing credentials or widening network access. ### What exactly did Cloudflare and Anthropic launch? Cloudflare’s May 19 press release described the product as “Cloudflare Environments for Claude Managed Agents,” available in partnership with Anthropic for organizations that want Claude to execute code, use tools and connect to private services on Cloudflare’s network. The company said customers can run the core agent loop on Anthropic’s Claude platform while using Cloudflare’s Workers developer platform to execute code and secure private connections. (cloudflare.com) Anthropic’s May 19 product update described the underlying feature as self-hosted sandboxes for Claude Managed Agents, with managed providers including Cloudflare, Daytona, Modal and Vercel. Anthropic said self-hosted sandboxes are in public beta, while MCP tunnels — a separate feature for private tool connectivity — are in research preview and require access requests. ### Why are sandboxes central to this release? (cloudflare.com) Anthropic said Claude Managed Agents can read files, run commands, browse the web and execute code, which makes the execution environment a security boundary as much as a compute layer. The company said self-hosted sandboxes let customers keep the agent’s execution environment within their own infrastructure choices and security controls, instead of relying only on Anthropic-hosted runtime. (claude.com) VentureBeat reported on May 19 that the design is aimed at moving credential control to the network boundary, so agents can reach enterprise APIs without carrying long-lived credentials inside the agent runtime itself. That account matches Cloudflare’s description of private service connectivity and isolated execution environments for untrusted code. (claude.com) ### What does Cloudflare add beyond raw compute? Cloudflare said its environment gives developers isolated execution, private service connectivity, logs and metrics, and the ability to customize sandbox images. In a company blog post, Cloudflare said developers can choose between traditional microVM-style isolation and lighter-weight isolates, depending on workload and scale requirements. The same post said private internal services can be connected without exposing them to the public internet. (venturebeat.com) Cloudflare framed that as useful for builders who want autonomous code execution close to data, tools or internal systems while retaining network controls and observability. ### How does this fit into Anthropic’s broader agent rollout? Anthropic launched Claude Managed Agents earlier in 2026 and has been adding enterprise controls around how those agents run and connect to outside systems, according to company materials and coverage published in May. (blog.cloudflare.com) InfoQ reported that Anthropic discussed Managed Agents and related developer tooling at its “Code with Claude 2026” event, while Anthropic’s own update on May 19 added self-hosted sandboxes and MCP tunnels. Cloudflare’s integration is one of the first named managed-environment options attached to that rollout. Anthropic’s documentation lists Cloudflare alongside Daytona, Modal and Vercel as providers that can handle compute and isolation for the sandbox layer. ### What can developers do next? Cloudflare published a May 19 blog post showing how to integrate a Cloudflare-based environment with Claude Managed Agents and said the service is intended for developers building autonomous code and tool-using workflows. (infoq.com) Anthropic said self-hosted sandboxes are in public beta on the Claude platform now, and MCP tunnels remain in research preview through an access-request process. (claude.com) Cloudflare’s press release said the launch was announced from San Francisco on May 19. Anthropic’s next step for users seeking private tool connectivity is the MCP tunnels request-access path, while developers who want managed sandbox execution can use providers named in Anthropic’s update, including Cloudflare. (cloudflare.com) (blog.cloudflare.com)