MCP Server Commands

🚀 Tool for LLMs Requests

This tool is designed for making requests to LLMs, specifically for the Claude desktop application. I'm really impressed that Claude Sonnet 3.5 can intelligently utilize these two tools.

✨ Features

• run_command: Run a command, such as hostname, ls -al, or echo "hello world".
  • It returns the text content of STDOUT and STDERR.
• run_script: Run a script (e.g., fish, bash, zsh, python).
  • Enable the LLM to execute the code it writes.
  • The script is passed via STDIN.
  • run_script is equivalent to run_command with the script passed via STDIN.
  • Claude has creatively used this feature, for example, using cat as an interpreter to create new files.

⚠️ Important Note

Be cautious about what you ask the server to run! In the Claude desktop application, use Approve Once (instead of Allow for This Chat) so that you can review each command individually. Use Deny if you don't trust the command. Permissions are determined by the user running the server. Do not run commands with sudo.

🎥 Video Demo

💡 Prompts

Prompts are used for the content that users include in the chat history, which can be added through Zed commands (in its AI chat panel).

• run_command: Generate a message with the command output.

🛠️ Development

📦 Install Dependencies

npm install

🏗️ Build the Server

npm run build

🔄 Develop with Automatic Rebuild

npm run watch

🚚 Installation

To use it with the Claude desktop application, add the server configuration:

• On MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• On Windows: %APPDATA%/Claude/claude_desktop_config.json

Use the Published npm Package

It's published on npm as mcp-server-commands. Use this workflow.

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

Use the Local Build (Checked out from the repository)

{
  "mcpServers": {
    "mcp-server-commands": {
      // This is because of the shebang in index.js
      "command": "/path/to/mcp-server-commands/build/index.js"
    }
  }
}

📝 Logging

The Claude desktop application writes logs to ~/Library/Logs/Claude/mcp-server-mcp-server-commands.log.

By default, only important messages (i.e., errors) are logged. If you want to view more messages, add --verbose to the args when configuring the server.

By the way, logs are written via STDERR because this is what the Claude desktop application routes to the log file. In the future, I expect well - formatted log messages to be transmitted via STDIO and written to the MCP client (note: not the Claude desktop application).

🐞 Debugging

Since the MCP server communicates via stdio, debugging can be challenging. We recommend using MCP Inspector, which is available as a package script:

npm run inspector

The Inspector will provide a URL to access the debugging tools in your browser.