GitHub radar

Cloudflare Computer: Make a sandbox for your AI agent.

Cloudflare Computer SDK provides a virtual filesystem and code execution environment within Cloudflare Workers. It offers durable filesystems backed by SQLite in Durable Objects, multiple runtime environments (full Linux container with FUSE, just-bash in a Dynamic Worker, or a JS module), and an MCP example.

01cloudflare/computer 8.7kTypeScript

Cloudflare Computer is an SDK provided by Cloudflare that allows AI agents to have their own virtual computer. This virtual computer has a durable filesystem (SQLite Durable Object), as well as a code execution environment. There are three different ways to run this code (full linux container with FUSE, just-bash in a Dynamic Worker, or a JS module), each of which can access the same durable filesystem via one API call. In addition, the MCP example in this repo provides a way for agents to interact with the workspace directly (examples/mcp). This means that you can use your Claude Code or other agent directly within the workspace without writing any code. This is currently labeled as preview and not intended for production, but there are working examples and demos for all the environments in the examples directory, including a tutorial for how to write a markdown document and convert it to a PDF using pandoc in a container. Cloudflare Computer has received 8700 stars in less than 3 months.

Why a vibe-coder should care

If you want your AI agent to be able to execute code and work with files on the server, Cloudflare Computer can provide a sandboxed environment for them to do so without having to worry about containers or VMs. You can also use MCP to have your agent make use of the workspace directly.

How to install

Copy this and send it to your agent — Claude Code, Codex, any of them:

Deploy the worker-shell example from https://github.com/cloudflare/computer — follow the packages/computer README, deploy to Cloudflare Workers, and show me a working endpoint that runs a bash command. Ask if you need my Cloudflare credentials.
Open on GitHub