Vercel Sandbox commands now support stdin
Vercel Sandbox processes can now accept input while running, enabling stdin-based MCP servers and other interactive CLI workflows inside sandboxed agent tasks.
Vercel Sandbox now supports sending input to running commands through **stdin**. The TypeScript SDK uses `stdin: true` and `detached: true` with `writeStdin()` and `closeStdin()`; Python supports `subprocess.PIPE` and `stdin.send()`.
Use this to run stdio MCP servers such as agent-browser in a sandbox, or to pass data to an interactive command while it runs. Python also supports passing all input at once through `run_process()`.
Vercel Sandbox now supports sending input to running commands through **stdin**. The TypeScript SDK uses `stdin: true` and `detached: true` with `writeStdin()` and `closeStdin()`; Python supports `subprocess.PIPE` and `stdin.send()`. Use this to run stdio MCP servers such as agent-browser in a sandbox, or to pass data to an interactive command while it runs. Python also supports passing all input at once through `run_process()`. The TypeScript SDK can forward a Node.js readable stream, and Python’s `communicate()` closes stdin and collects output while waiting for exit. The material does not specify behavior for other SDKs or process-management limits.
This adds the missing input channel for interactive processes in Vercel Sandboxes, making it practical to run stdio MCP servers there or send data while a command is running. It clarifies the sandbox’s process interaction options, without establishing broader isolation or security guarantees.