g0t4/mcp-server-commands

Source Code

Description

A general-purpose Model Context Protocol (MCP) server that can execute any command with run_command and run_script tools, demonstrating flexible server capabilities under the MCP standard. This server is designed for automation and integration with LLMs, such as via the Claude Desktop app.

Features

• run_command Tool:
  • Execute arbitrary shell commands (e.g., hostname, ls -al, echo "hello world").
  • Returns both STDOUT and STDERR as text.
• run_script Tool:
  • Run scripts in various interpreters (e.g., fish, bash, zsh, python).
  • The script is passed over STDIN, making it suitable for letting LLMs run generated code.
  • Supports creative uses (e.g., using cat as an interpreter to create files).
• Prompts Integration:
  • Prompts can be included in chat history for integration with tools like Zed's slash commands.
  • Can generate prompt messages with command outputs.
• Installation & Usage:
  • Available as an npm package (mcp-server-commands).
  • Can be configured for Claude Desktop app on macOS and Windows.
  • Supports both npm-based and local builds.
• Logging:
  • Logs are written to STDERR and can be made verbose with a flag.
  • Claude Desktop writes logs to a dedicated file.
• Debugging:
  • MCP Inspector tool available via npm run inspector to provide debugging tools in the browser.
• Open Source:
  • Licensed under the MIT license.
  • Written in JavaScript and TypeScript.

Pricing

• Open source (MIT license); free to use.

Tags

command-execution, automation, mcp, open-source