# Vercel Sandbox commands now accept stdin input

published: 2026-10-26T18:05:55.127Z
topic: infrastructure
source: https://vercel.com/changelog/vercel-sandbox-commands-now-support-stdin
url: https://aiagentnews.io/stories/vercel-sandbox-commands-now-accept-stdin-input

Vercel Sandbox commands can now receive input through standard input (stdin) while they are running. The feature is available in both the TypeScript and Python Sandbox SDKs.

Vercel Sandbox now lets callers send input to a running command through standard input (stdin). Vercel names two uses: exchanging messages with stdio MCP servers such as agent-browser, and sending data to other command-line programs while they run.

In the TypeScript SDK, developers set stdin and detached to true in runCommand(), then call writeStdin() to send input and closeStdin() when they are finished. A Node.js readable stream, such as process.stdin, can also be passed as stdin, and the SDK forwards its data and closes the command's input when the stream ends.

In the Python SDK, the process is created with stdin=subprocess.PIPE and input is sent through server.stdin.send(). Calling communicate() closes stdin, collects the remaining output and waits for the process to exit. Commands that receive all their input at once can instead get it passed directly to run_process().

## Why it matters if you build agents

Agents that work with stdio MCP servers inside a sandbox can now keep an ongoing exchange with the process instead of starting it with fixed input. Code that streams data into a command should close stdin, using closeStdin() in TypeScript or communicate() in Python, so the command receives end-of-input and exits.