# mcp-server-commands

An MCP server with a single runProcess tool that lets an LLM run shell commands or direct executables on the host machine, returning stdout and stderr, with an optional stdin channel for scripts or file creation.

This MCP server exposes one tool that runs a command on the host either through the system's default shell, so pipes, redirects, and variable expansion work, or as a direct executable invocation with an argument array and no shell interpretation; the caller picks one form or the other, not both. An optional stdin parameter lets the model pass a script to an interpreter such as bash, fish, or python, or write a file with a command like an append redirect.

It also defines a matching prompt for use in chat history, for example through editor slash commands, so a command's output can be inserted as a message rather than only returned as a tool result.

It installs with standard npm commands, and is configured in Claude Desktop's server config on macOS or Windows. Because the tool can run arbitrary commands with the user's own permissions, the README recommends approving each command individually rather than allowing it for an entire chat, and warns against running the server with elevated privileges.

Website: https://github.com/g0t4/mcp-server-commands