MCP Ssh Manager

What bvisible/mcp-ssh-manager MCP server does

bvisible/mcp-ssh-manager MCP server exposes SSH-based administration capabilities to an MCP client. It is designed for managing several remote servers from one configuration rather than connecting to each host manually. Supported tasks include running commands, moving files, synchronizing directories with rsync, creating and restoring backups, checking service and host health, and maintaining persistent remote sessions.

Database operations cover MySQL, PostgreSQL, and MongoDB. The listed functions include database dumps, imports, and queries. Read-only SQL enforcement is applied to ssh_db_query, which rejects statements that are not SELECT queries.

The server also supports SSH tunnels, ProxyJump, bastion hosts, and optional per-server SSH agent forwarding. Agent forwarding requires a running local SSH agent and does not copy private keys to the remote host.

How it works

The MCP server uses the OpenSSH client available on the host where it runs, and can target SSH servers on Linux, macOS, and Windows. Server definitions can include an optional group, allowing group-based command execution. Groups may be defined in .env or TOML configuration and can be combined with a .server-groups.json file.

Access policy is configured separately for each server. unrestricted is the default mode. readonly blocks mutating actions such as deployment, uploads, sudo operations, and database imports while retaining read operations. restricted requires every command to match an allow pattern and not match a deny pattern before it is sent to the host.

Sudo passwords are sent through the SSH channel’s standard input instead of being placed in the remote command line. The project also centralizes shell quoting for database, backup, and other command builders, with tests covering injection payload combinations.

Setup and configuration

The project is published as the npm package mcp-ssh-manager. The supplied material does not include a complete installation command or a full configuration schema, so those details should be taken from the repository’s current documentation before deployment.

Configuration examples use per-server environment-variable names such as SSH_SERVER_PROD_MODE, SSH_SERVER_STAGING_MODE, and SSH_SERVER_STAGING_ALLOW_PATTERNS. Server settings can also include a group value and the forward_agent or FORWARD_AGENT option. Agent forwarding is opt-in and defaults to disabled.

Because the tool can execute commands on important hosts, configure security modes per server before giving an agent access. Review where the MCP host captures stderr and logs, and protect the SSH credentials and configuration used by the process.

Tools and capabilities

bvisible/mcp-ssh-manager MCP server supports capabilities in these areas:

• Execute commands on named SSH servers or server groups.
• Transfer files and synchronize directories with rsync.
• Run MySQL, PostgreSQL, and MongoDB dump, import, and query operations.
• Create and restore backups.
• Inspect service status and host health, including tailing logs.
• Maintain persistent SSH sessions.
• Create SSH tunnels and connect through ProxyJump or bastion hosts.
• Apply read-only or restricted command policies per server.
• Record an audit log and optionally forward the local SSH agent.

Limitations and notes

The default unrestricted mode permits the full operation set, so it should not be treated as a safety boundary. Restricted patterns must be designed carefully, and readonly mode does not mean the agent has no access to remote data.

The README identifies security fixes in version 3.8.5 affecting backup tools, database dumps, service-status checks, log tailing, and the readonly or restricted policy path. Users relying on those features should upgrade to the fixed release and review the project’s security advisories. The provided material does not specify supported MCP clients beyond Claude Code and OpenAI Codex, nor does it document an authentication service or hosted endpoint.