The linux-mcp-daemon (mcpd) is a Go-based tool designed to manage Linux servers via the Model Context Protocol (MCP). It allows AI agents to interact with system resources such as processes, disks, and services using a standardized JSON-RPC interface. The daemon provides 38 tools and 11 permanent resources, enabling granular control without requiring full SSH access.
Unlike standard shell-based MCP implementations, mcpd operates as a network service using HTTPS. It avoids the security risks associated with stdio-based transports by running each tool execution as a short-lived worker process under the specific Linux user account. This ensures that the AI agent's permissions are restricted by the underlying operating system's access controls.
The system includes a configuration file, mcp-sudo.yaml, which defines which users can execute specific tools with root privileges. It supports path-based restrictions for file operations and network access controls. Every action performed by the agent is logged, with modifications to the system marked as audit events. The daemon generates self-signed TLS certificates by default for secure communication.
The project includes a command-line client, linuxctl, which functions similarly to kubectl. It retrieves tool schemas from the daemon in real-time, supporting features like auto-completion and resource inspection. The tool is available via installation scripts, .deb and .rpm packages, or as a Docker container, and is compatible with AI agents like Claude Code.
