Mcp Remote Ssh

Name: Mcp Remote Ssh
Author: faizbawa

@faizbawa

About Mcp Remote Ssh

MCP server giving AI agents full SSH access with secret-safe environment variable injection. Supports persistent sessions, structured command execution, SFTP file transfer, port forwarding, and automatic output redaction that prevents secrets from reaching the LLM.

#ssh #remote-access #devops #secrets #sftp #paramiko #ai-agents #mcp #security

Basic information

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "remote-ssh": {
      "command": "uvx",
      "args": [
        "mcp-remote-ssh"
      ]
    }
  }
}

Tools

Connect to a remote host via SSH. Returns a session_id for use with all other tools. Supports password and key-based authentication.

List all active SSH sessions with their connection status and details. Returns: List of session info dicts.

Close an SSH session and release all its resources (shell, SFTP, port forwards). WARNING: this kills any running processes in the session.

Execute a command on the remote host and return structured output. Each call runs in an independent exec channel -- no state is shared between calls (use ssh_shell_* tools for persistent state). If secrets have been loaded via ssh_load_env_file, they are automatically injected as environment variables for this command.

Execute a command with sudo on the remote host. If the user already has passwordless sudo, leave sudo_password empty. If secrets have been loaded via ssh_load_env_file, they are automatically injected as environment variables for this command.

Create an SSH port forward (local -> remote). Connections to 127.0.0.1:local_port will be tunneled through SSH to remote_host:remote_port. If local_port is 0, a random available port is chosen.

List all active port forwards for an SSH session.

Close a specific port forward.

Load environment variables from a LOCAL file (on the machine running this MCP server) into the remote SSH session. The file is read from the local filesystem -- it does NOT need to exist on the remote host. Variable VALUES are never returned to the caller; only variable names are confirmed. Loaded values are: 1. Registered for automatic redaction -- any subsequent tool output (ssh_execute, ssh_shell_send, ssh_shell_read, ssh_read_remote_file) that contains a secret value will have it replaced with '***'. 2. Exported into the remote shell (if open) via builtins, avoiding exposure in the process tree on the remote host.

Clear all loaded secrets from the redaction registry. Secrets will no longer be redacted from tool output. Does NOT unset the environment variables from the remote shell.

Upload a local file to the remote host via SFTP.

Download a file from the remote host to the local machine via SFTP.

Read a text file on the remote host and return its contents. For large files, use max_bytes to limit the amount read.

Write text content to a file on the remote host via SFTP.

List files and directories at a path on the remote host via SFTP.

Open a persistent interactive shell on the SSH session. The shell preserves working directory, environment variables, and running processes across multiple send/read calls. Ideal for screen/tmux, long builds, etc. If a shell is already open, this is a no-op (returns existing shell info). If secrets have been loaded via ssh_load_env_file, they are automatically injected into the new shell.

Send text to the interactive shell. By default appends Enter (newline) and waits briefly to capture output.

Read the current content of the interactive shell buffer. Use this to poll for output from long-running commands without sending anything.

Send a control character to the interactive shell. Common keys: "c" for Ctrl+C (interrupt), "d" for Ctrl+D (EOF), "z" for Ctrl+Z (suspend), "l" for Ctrl+L (clear screen), "a" for Ctrl+A (screen prefix).

Wait for the shell output to contain a specific pattern, or for the output to stabilize (no new output for two poll intervals). Useful for waiting on long-running commands to complete.