MCP server exposing SSH control for Linux and Windows systems via Model Context Protocol.
SSH MCP has 6 tools with basic schemas and descriptions, but significant gaps limit production readiness. All tools have descriptions (10-60 chars, mostly short), and input schemas are present with typed parameters. However, descriptions lack depth, they state WHAT but not WHEN or WHY to use each tool, and lack dependency hints. No output schemas are documented. Error handling is minimal: tools return McpError for validation but lack recovery guidance or actionable error messages. Parameter descriptions are sparse (e.g., 'The absolute path to the file to read' is minimal; no format constraints, no examples of valid paths). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY vs WRITE). Password tools (set_su_password, set_sudo_password) expose security concerns: storing passwords in memory without clear lifecycle or rotation guidance. Overall, the server is functional but lacks the depth and safety guardrails expected of production tools.
Execute a shell command on the remote SSH server
Read the contents of a file from the remote SSH server
List the contents of a directory on the remote SSH server
Write contents to a file on the remote SSH server
Set the su (switch user) password for privilege elevation
Set the sudo password for privilege elevation
No output schemas documented. LLMs cannot plan downstream calls or extract required fields (e.g., what does execute_command return? stdout, stderr, exit code, all three?). Breaks tool chaining.
Descriptions are too short (10-60 chars) and lack context. 'Execute a shell command on the remote SSH server' does not explain WHEN to use this vs other tools, what happens on failure, or whether it's idempotent. Descriptions should be 50-200 chars with WHAT, WHEN, and prerequisites.
Password tools (set_su_password, set_sudo_password) accept passwords as string parameters. This violates secret-injection pattern: credentials in parameters are logged and leak into traces. Passwords should be injected server-side via environment variables or vault, not exposed as tool inputs.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. execute_command, put_file, set_su_password, set_sudo_password are destructive/write operations; get_file, list_directory are read-only. Annotations help LLMs reason about safety and retry logic.
Error handling lacks recovery guidance. sanitizeCommand() returns 'Command is too long' but does not suggest alternatives (e.g., 'Split into multiple commands' or 'Increase --maxChars limit'). Errors should guide the LLM's next step per pattern:recovery-guide.