bvisible/mcp-ssh-manager

What it solves

This project provides a secure way for AI agents (like Claude Code and OpenAI Codex) to manage remote servers via SSH. It bridges the gap between an AI's reasoning and the actual execution of DevOps tasks, allowing an agent to perform server management, file transfers, and database operations without requiring the user to manually run every command.

How it works

It acts as a Model Context Protocol (MCP) server that exposes a set of tools to an AI assistant. The assistant can then call these tools to execute commands, manage databases, and monitor health across multiple configured servers. To ensure safety, it implements a granular security layer where users can define access levels per server:

• Unrestricted: Full access.
• Read-only: Blocks mutating tools (like uploads or sudo) and applies a denylist to shell commands.
• Restricted: A "fail-closed" mode where only commands matching specific allow-patterns are permitted.

Who it’s for

DevOps engineers, system administrators, and developers who want to integrate their AI assistants into their server management workflow while maintaining strict security controls over what the AI can actually execute on their infrastructure.

Highlights

• Granular Security: Per-server security modes (readonly, restricted, unrestricted) to prevent AI-driven accidents or malicious injections.
• Secure Credential Handling: Sudo passwords are sent via the SSH channel's stdin rather than the command line to avoid leaking in process lists.
• Infrastructure Support: Supports ProxyJump/Bastion hosts, SSH agent forwarding, and Windows OpenSSH hosts.
• Database Integration: Includes specialized tools for database queries and dumps with enforced read-only SQL for specific tools.
• Comprehensive Toolset: 37 different tools for tasks ranging from health monitoring and backups to file synchronization via rsync.