Cloudflare Sandboxes Bring Cursor Cloud Agents to Secure Environments

Author : John Brown | Published On : 04 Sep 2026

Cloudflare is expanding its support for AI-powered development tools by bringing Cursor Cloud Agents to Cloudflare Sandboxes. The integration gives developers and enterprise teams a way to run AI coding agents inside secure, customer-controlled environments while continuing to use the Cursor workflow they already know. Announced on September 2, 2026, the move highlights the growing demand for AI agents that can perform development tasks without requiring organizations to surrender control over their code, systems, and secrets.

Cloudflare Extends Sandboxes to Cursor Cloud Agents

AI coding agents are becoming increasingly capable of handling tasks such as writing code, testing applications, navigating files, and interacting with development environments. However, giving autonomous agents access to source code and internal systems creates new security and infrastructure challenges.

Cloudflare’s latest integration addresses part of this challenge by allowing Cursor Cloud Agents to use Cloudflare Sandboxes as their execution environment. Instead of running agent tasks directly on a developer’s local machine or on infrastructure outside an organization’s control, teams can route the workload to a sandbox environment within their Cloudflare account.

Cloudflare describes Sandboxes as isolated environments designed for AI agents and developer tools. Each sandbox provides a dedicated execution environment where code can be run, files can be managed, packages can be installed, and development services can operate separately from the underlying infrastructure.

Cursor Keeps the Existing Developer Workflow

One of the key benefits of the integration is that developers do not have to abandon the Cursor experience they already use.

Cursor Cloud Agents can be launched and managed through Cursor's desktop application, website, or mobile application. The Cursor platform continues to handle the agent loop, including inference, planning, and orchestration, while the actual execution of tasks can take place on a customer-selected worker.

With Cloudflare Sandboxes serving as that worker environment, terminal commands, filesystem operations, browser actions, and other agent activities can run within an isolated environment controlled by the customer.

This creates a separation between the intelligence driving an agent and the infrastructure where its work is performed. Developers can continue interacting with Cursor while organizations gain greater control over the environment in which agent workloads execute.

Why Customer-Controlled Execution Matters

As AI agents become more autonomous, they increasingly need access to development resources. An agent may need to clone a repository, install dependencies, run tests, start a development server, inspect files, or interact with a browser.

Those capabilities are useful, but they also introduce security concerns when agents operate on sensitive code or within corporate environments.

Cloudflare Sandboxes are designed to provide isolated execution for these workloads. Cloudflare says each sandbox runs in its own secure container and provides an isolated filesystem. Sandboxes can also maintain state while the container remains active, allowing them to support more complex development tasks.

For enterprises, this approach can make it easier to establish boundaries around AI-generated or AI-executed workloads while maintaining existing development processes.

Self-Hosted Machines Provide the Connection

The integration works through Cursor's Self-Hosted Machines model. In this architecture, a Cursor worker runs the Cursor CLI and establishes a long-lived outbound HTTPS connection with Cursor's backend.

This outbound connectivity model means organizations do not need to open an inbound connection from Cursor into their private network. Instead, the customer-controlled worker establishes the connection and receives the relevant agent tool calls through it.

Cloudflare Sandboxes can serve as the customer-controlled worker environment, giving teams another infrastructure option for running Cursor Cloud Agent workloads.

This model is particularly relevant for companies that want to maintain control over where repositories, build caches, and other development resources are located.

Code, Build Caches and Secrets Stay Under Customer Control

Control over sensitive development resources is one of the major considerations behind customer-managed agent execution.

According to Cloudflare, when Cursor operates through self-hosted machines, repositories, build caches, and secrets remain on the customer's machines. Some information, such as file chunks accessed by the model during inference and Cloud Agent artifacts including screenshots, videos, and log references, can still be uploaded so they can appear in dashboards and pull requests.

This distinction allows organizations to keep core development resources within their selected execution environment while still benefiting from the centralized Cursor experience.

For teams with strict requirements around source-code location, credentials, or internal infrastructure, this can provide an additional layer of operational control.

Cloudflare Positions Sandboxes as an Agent Execution Layer

The Cursor integration is part of Cloudflare's broader strategy around infrastructure for autonomous AI agents.

Cloudflare has been expanding its Agent Cloud and developer platform with tools designed to help developers build, deploy, and operate AI agents at scale. Its Sandboxes platform is specifically positioned as a secure execution environment for agents that need to write and run code.

The company has also integrated its sandbox infrastructure with other AI agent platforms. In May 2026, Cloudflare announced an integration with Anthropic's Claude Managed Agents, while its platform also supports Devin Outposts.

Adding Cursor to this ecosystem strengthens Cloudflare's position as an infrastructure provider for agent workloads rather than simply an application platform.

Sandboxes Built for AI Development Workloads

Cloudflare Sandboxes are designed around the needs of software agents and developer tools.

The platform allows developers to execute commands, manage files, run background services, work with Git repositories, and expose services through controlled environments. Cloudflare also provides support for code interpretation and other development-oriented workloads.

The underlying sandbox architecture is built on Cloudflare Containers, providing global placement, automatic scaling, and usage-based infrastructure. Cloudflare also highlights fast sandbox startup times, allowing developers to create execution environments without maintaining large pools of idle infrastructure.

These capabilities are important for AI agents because workloads can be highly dynamic. An agent may need significant compute for a short period while completing a coding task and then stop using the environment once the task is finished.

What the Integration Means for Enterprise Development

For enterprise development teams, the announcement represents a shift toward more flexible agent deployment models.

Organizations increasingly want to use AI coding assistants but may have concerns about allowing autonomous systems to operate directly inside production or corporate environments. Customer-controlled execution environments can help organizations establish clearer boundaries for these workloads.

With Cursor Cloud Agents running through Cloudflare Sandboxes, teams can retain the familiar Cursor interface while choosing where the agent performs its actual work.

This could be particularly valuable for organizations managing large engineering teams, sensitive intellectual property, regulated workloads, or distributed development infrastructure.

Part of a Broader Shift in AI Agent Infrastructure

The Cloudflare-Cursor integration reflects a broader change in how AI agents are being deployed.

Early AI coding tools primarily assisted developers inside local editors. Newer agentic systems can operate for longer periods, execute multiple steps, interact with tools, and make changes across development environments.

As these capabilities expand, the execution environment becomes just as important as the AI model itself.

Companies need infrastructure that can provide isolation, networking, file access, observability, scalability, and security without forcing developers to completely change their workflows.

Cloudflare is positioning Sandboxes to address this infrastructure layer. Its existing integrations with platforms such as Claude Managed Agents and Devin Outposts, followed by support for Cursor Cloud Agents, demonstrate the company's broader focus on becoming an execution platform for agentic applications.

The Future of Customer-Controlled AI Agents

The arrival of Cursor Cloud Agents on Cloudflare Sandboxes gives developers another way to balance AI-assisted productivity with infrastructure control.

Rather than treating AI agents as tools that must operate entirely inside a vendor-controlled environment, the emerging model allows the agent's intelligence and execution environment to be separated. Developers can use familiar AI tools while organizations decide where code execution occurs and what resources agents can access.

For Cloudflare, the latest integration further expands its role in the rapidly developing AI infrastructure market. For Cursor users, it provides another option for running cloud-based coding agents on customer-controlled infrastructure.

As AI coding agents take on increasingly complex software development tasks, secure and flexible execution environments are likely to become a fundamental part of enterprise AI adoption.

Discover IT Tech News for the latest updates on IT advancements and AI innovations.