PermShell MCP is an MCP server that enables AI assistants to execute shell commands with permission notifications. Designed for software engineers, system administrators, and developers using MCP-compatible clients like Claude Desktop, it connects local terminal capabilities directly to large language models while maintaining human oversight. Rather than running commands automatically in the background, PermShell MCP integrates transparent permission dialogs via growl notifications that display the exact command and its target working directory before execution begins. Users can review, approve, or reject command execution in real time. The server also includes built-in safeguards such as execution timeout limits to prevent runaway tasks and input sanitization to defend against command injection. In addition to shell execution, it exposes dedicated utilities to inspect host system resources including CPU, memory, storage disks, uptime, and operating system details.
Category: Cloud & Infrastructure
Tags: bash, command-execution, security, shell, terminal
Follow these steps to install and register PermShell MCP with Claude Desktop: 1. Clone the repository and navigate into the directory: bash git clone https://github.com/mcollina/perm-shell-mcp.git cd perm-shell-mcp 2. Install dependencies: bash npm install 3. Build the server: bash npm run build 4. Open your Claude Desktop configuration file and add the server definition under mcpServers: json { "mcpServers": { "permshell": { "command": "node", "args": ["/path/to/perm-shell-mcp/dist/index.js"] } } } Replace /path/to/perm-shell-mcp with the absolute path to your built repository.
Part of MCP Servers
PermShell MCP is a Model Context Protocol server that allows LLMs to execute command-line operations on your host system. It includes explicit confirmation dialogs using desktop growl notifications to ensure users verify every command before it runs.
It enforces multiple safeguards including explicit permission requests via desktop notifications, display of the exact working directory, automatic timeouts to stop runaway processes, and input sanitization to guard against command injection.
The server exposes two main tools: execute-command, which runs an approved shell command on the host, and system-info, which returns hardware and operating system status such as CPU, disk, memory, and uptime.
You clone the GitHub repository, install dependencies using npm install, run npm run build, and then reference the generated dist/index.js file using Node inside the Claude Desktop configuration file under mcpServers.