Cloudflare announces that developers can now run Cursor Cloud Agents in Cloudflare Sandboxes. This integration gives companies and platform teams more control over where AI agents operate.
Cloudflare is expanding its support for AI agent platforms with the ability to run Cursor Cloud Agents in Cloudflare Sandboxes. This allows organizations to run AI coding agents in a self-managed, secure environment. The announcement builds on previous integrations with, among others, Devin Outposts and Claude Managed Agents.
Integration with existing AI agent platforms #
Cloudflare’s collaboration with Cursor follows previous steps regarding other AI agent platforms, such as Devin Outposts and Claude Managed Agents. The goal is to offer developers flexibility in their workflow without companies losing control over the agent’s work. With the new integration, developers can initiate coding tasks from the Cursor app, web interface, or mobile application. The actual agent work is performed on machines managed by the customer themselves, separate from the serverless Cloudflare Workers platform.
Cursor Self-Hosted Machines remain responsible for executing the agent loop, such as inference, planning, and orchestration. The actual tasks are performed on a worker selected by the customer, often within a sandbox environment in the Cloudflare account.
Management and security #
With the arrival of Cursor Cloud Agents on Cloudflare Sandboxes, developers can maintain their existing Cursor workflow. The difference lies in the execution location: actions such as terminal and file system operations now occur in isolated, customer-managed environments. This is particularly relevant for organizations with strict compliance requirements regarding code and data.
Teams have the option to link individual workers or create worker pools for different execution environments. Pool orchestration ensures that capacity can be scaled up or released depending on the workload. During self-hosting, repositories, build caches, and secrets remain on the local machines, while relevant artifacts such as screenshots and logs can be uploaded for further processing in dashboards and pull requests.