MCP Server Commands

What g0t4/mcp-server-commands MCP server does

g0t4/mcp-server-commands MCP server exposes one MCP tool, run_process, for starting processes on the machine where the server runs. It is intended for command-line tasks requested through an MCP-compatible client, such as checking the hostname, listing files, printing text, or invoking an installed interpreter.

The tool returns the process's STDOUT and STDERR as text. An optional stdin value lets the model provide input to programs such as bash, fish, zsh, or python. It can also support commands that write incoming text to a file, provided the command itself performs that operation.

How it works

The tool accepts one of two mutually exclusive command formats:

• command_line: a single string executed by the system's default shell. Shell behavior such as pipes, redirects, and variable expansion is available.
• argv: a string array where the first item is the executable and later items are its arguments. This invokes the executable directly without shell parsing.

The selected parameter determines whether the server uses shell execution or direct execution. If a particular shell should be used, the README recommends specifying that preference in the model's system prompt or tool instructions. The server uses the MCP STDIO transport.

Setup and configuration

The published npm package can be launched through npx in an MCP client configuration:

{
  "mcpServers": {
    "mcp-server-commands": {
      "command": "npx",
      "args": ["mcp-server-commands"]
    }
  }
}

For a repository checkout, build the project with npm install followed by npm run build, then point the client at build/index.js. npm run watch is available for automatic rebuilds during development. Claude Desktop stores its configuration under the application support directory on macOS or the application data directory on Windows. Groq Desktop's documented configuration path is also provided for macOS.

The package supports a --verbose argument in the client configuration for more detailed logging. The server writes logs to standard error; Claude Desktop routes them to its MCP log file. npm run inspector starts the MCP Inspector workflow for debugging.

Tools and capabilities

g0t4/mcp-server-commands MCP server currently provides:

• run_process for shell-based or direct process execution.
• Text results containing standard output and standard error.
• Optional standard input for scripts and other commands.
• Prompt support that can generate a prompt message containing command output, including use from Zed's AI Chat slash commands.

The example connects that bridge to Open WebUI.

Limitations and notes

g0t4/mcp-server-commands MCP server runs with the permissions of the user account that starts it. It can therefore expose the host to risky or destructive commands. Review every requested command, use per-invocation approval where the client provides it, and do not run the server with sudo.

Direct argv execution avoids shell interpretation, while command_line intentionally permits shell features. Local models may also need system-prompt guidance before they will call the tool. When using the mcpo bridge or Open WebUI, review the security implications before exposing the interface.