Give Claude, ChatGPT or Cursor its own cloud computer.
Add https://sandboxer.com/mcp as a connector and sign in. The assistant can then create Linux sandboxes, run commands, move files and put a port on an HTTPS URL, all in microVMs, not on your machine.
Nothing to install
The hosted server speaks Streamable HTTP. Apps sign in with your Sandboxer account over OAuth; headless clients send an API key.
Docs20 tools for the whole lifecycle
Create, list, rename and label sandboxes, run commands in the foreground or background, read and write files, host ports, save snapshots, stop and delete.
DocsShow the user a running app
host_port puts a port on an HTTPS URL, so the assistant can hand you a link to the app it just built.
DocsLocal too
sbx mcp runs the same tools as a stdio server inside the CLI, for clients that prefer a local process.
DocsAdd it to Claude Code
claude mcp add --transport http sandboxer https://sandboxer.com/mcp
# or run it locally from the CLI
claude mcp add sandboxer -- sbx mcpMore use cases
Give every agent its own Linux VM with your repo, your keys and a real Docker daemon. Start ten from the same template, prompt them over the API, and keep the ones that worked.
Start a sandbox from your app's template when a PR opens, check out the branch, and post a stable HTTPS link. Reviewers click it; it stops when the PR closes.
Every sandbox has a Linux desktop with Chrome that an agent can see and drive, and that you can watch live in a browser tab.
When your product runs code an LLM wrote or a user uploaded, run it in a VM with its own kernel that can't reach your network, then throw it away.
/dev/kvm is passed through to every sandbox, so QEMU, Firecracker and the Android emulator run with hardware acceleration inside it.
Leave a build, a crawl or an agent running in a sandbox. It pauses with its memory when it goes idle and wakes on the next request, so nothing restarts from scratch.