What it does
Safe SSH MCP Server gives an AI agent a fixed set of read-only system diagnostic commands on a remote host over SSH, instead of an unrestricted shell. The commands cover disk and filesystem usage, CPU, memory and process state, systemd units and logs, kernel messages, listening sockets, routes and DNS lookups, TCP reachability, Docker and Kubernetes containers and events, OS and installed package details, and failed logins and core dumps. The aim is to prevent changes to the remote system; it does not control what an external agent does with the output.
It suits an engineer or operator who wants an agent to inspect a Linux server, for example to check which directory uses the most disk space, review failed systemd units, or read recent warning-level journal entries, without giving that agent write access to the host.
The project is written in Python and is tested on Python 3.11 against a remote Ubuntu server, with automated tests on Python 3.11 to 3.14. Authentication can use an account password or a private key. For passphrase-protected keys, the source offers an environment variable or, on macOS, the Keychain and SSH agent, which keeps the passphrase out of the server's configuration. The project includes health-check scripts for the SSE and stdio transports and is licensed under AGPLv3.
Limitations and requirements
Tested manually on macOS with Python 3.11 against Ubuntu; automated tests cover Python 3.11 to 3.14. Account-password or private-key login is required. The KEYCHAIN option is for macOS only and needs the key unlocked in the same user session that starts the server. The ENV_VAR option stores the passphrase in the server process environment, so access to that process must be restricted. Putting a passphrase directly in a shell command may save it in shell history. The 'safe' label refers to host safety only, not to protecting the output from misuse by external agents. The project is licensed under AGPLv3.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 17 Nov 2026 · its page on the source
io.github.Areso/safe-ssh-mcp
A secured scoped SSH MCP server for executing safe read-only diagnostic DevOps / SysOps commands
v0.2.3As shown on registry.modelcontextprotocol.io when we last read it