使用说明
Cloudflare Container Sandbox MCP Server
This is a Model Context Protocol (MCP) server that supports remote MCP connections, with Cloudflare OAuth built-in.
It integrates tools for running a sandbox container with your MCP client. With this server you can allow your LLM to run arbitrary code, such as Node or Python, in a secure, sandboxed environment.
The /mcp URL uses a stateless SDK v2 handler that creates a fresh server for every request. The retired /sse URL returns 410 Gone pointing at /mcp. UserContainer and ContainerManager remain application Durable Objects for per-user container lifecycle and capacity management; there is no MCP protocol session or protocol Durable Object.
Each user's sandbox is a Container controlled by their UserContainer Durable Object through the native ctx.container API, with the durable_object scheduling policy. Every tool runs as a process with ctx.container.exec(); the container itself runs no server.
Sandboxes start from a snapshot of the Cloudflare-managed cloudflare/debian-trixie image with Python, pip, git, build tools and pnpm installed. One UserContainer instance, __sandbox-template__, builds that snapshot the first time a sandbox starts and keeps the handle (server/sandboxTemplate.ts). Every restore refreshes the snapshot's 30-day lifetime; if a restore fails, for example because the snapshot expired, the template is rebuilt.
Tools
| Category | Tool | Description |
|---|---|---|
| Container Lifecycle | container_initialize |
(Re)start a container. Containers are intended to be ephemeral and don't save any state. Containers are only guaranteed to last ~10m. |
container_ping |
Ping a container for connectivity | |
| Filesystem | container_file_write |
Write to a file |
container_files_list |
List all files in the work directory | |
container_file_read |
Read the contents of a single file or directory | |
container_file_delete |
Delete a single file or directory | |
| Execution | container_exec |
Run a command in the shell |
This MCP server is still a work in progress, and we plan to add more tools in the future.
Prompt Examples
Create a visualization using matplotlib. Run it in the container that you can startClone and explore this github repo: [repo link]. Setup and run the tests in your development environmentAnalyze this data using Python
Connect to the MCP server
Connect your MCP client directly to https://containers.mcp.cloudflare.com/mcp. If prompted, complete the Cloudflare OAuth flow in your browser. The tools become available after authorization.
Interested in contributing, and running this server locally? See CONTRIBUTING.md to get started.
