# SSH MCP Server

SSH MCP Server gives AI agents controlled, audited SSH access to remote Linux hosts, classifying each command and requiring policy-based or human approval before anything destructive runs.

SSH MCP Server is a Model Context Protocol server that lets an LLM agent run commands over SSH on configured remote hosts. It classifies each command before execution, authorizes it against a role-by-host-group matrix, can require human approval for destructive actions, and logs every decision either way.

The server starts even without a configuration file, so a client or directory can complete the MCP handshake and list its tools, but every tool call is refused until a config file is created at a per-platform path. On Linux and macOS the server checks that only the file's owner can read the configuration file and its directory, and refuses to start otherwise; Windows enforcement differs because file permissions there answer that question less directly.

It is written in TypeScript and installed from npm.

Website: https://github.com/tufantunc/ssh-mcp