This MCP server implements a focused tool for LLM-driven automation: the run_command interface. It executes arbitrary shell commands on the host process and returns both STDOUT and STDERR as text. An optional stdin parameter lets the LLM feed code or input to interactive shells (e.g., bash, zsh, fish, python) and even perform file operations (for example via cat >> foo/bar.txt). The server is designed to be used with Claude Desktop and Groq Desktop, acting as a bridge between the LLM and the host environment while keeping permissions under the user’s control (no sudo). It uses the STDIO transport for command communication, and can be exposed over HTTP via mcpo to provide an OpenAPI-compatible web server interface, compatible with Open-WebUI ecosystems. Development workflows are npm-based (npm install, npm run build, npm run watch for auto-rebuild). Logging is available and can be increased with verbose mode; logs are written to STDERR for compatibility with Claude Desktop routing. Debugging can be facilitated through the MCP Inspector. The README also provides configuration guidance for local deployment and HTTP bridging.