SSH Model Context Protocol Server - Enable SSH functionality for AI models
This SSH MCP server has significant definition quality gaps. Tool schemas are present and mostly complete with proper types and descriptions, but there are critical issues with naming clarity, parameter documentation inconsistency, and lack of structured output schemas. The server registers 9 tools with basic input schemas visible in ssh_mcp/server.py, but the tool descriptions are generic and lack the specificity required for LLM reasoning. Several tools have ambiguous names and missing parameter constraints. The descriptions do not explain when to use each tool vs alternatives (e.g., ssh_connect vs ssh_config), prerequisites, or return structures. Parameters like 'action' in ssh_key_manage are under-constrained (accepts string without enum). No evidence of documented output schemas or pagination support. Error handling is not visible in the provided code. The server relies on session_id tracking, which requires multi-step planning (connect first, then exec/upload/download), but tools do not guide this dependency chain.
Load SSH configuration from file
Connect to a remote SSH server
Disconnect from a remote SSH server
Download a file from a remote SSH server
Execute a command on a connected SSH server
Manage SSH keys
List all active SSH sessions
Create SSH local port forwarding tunnel
Parameter constraints under-specified: ssh_key_manage 'action' accepts free-form string instead of enum ['add', 'remove', 'list']. LLM will hallucinate invalid actions.
Output schemas not documented: No explicit schema for return values of any tool. LLMs cannot plan downstream tool calls or extract fields reliably. E.g., ssh_connect returns session_id but this is inferred, not declared.
Tool descriptions are generic and under 50 chars: Descriptions lack context on prerequisites, when to use each tool vs alternatives, and return structures. E.g., ssh_connect and ssh_config both relate to configuration, no guidance on which to use when.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 53 | - | v1 |
Upload a file to a remote SSH server
Multi-step tool dependencies not documented: ssh_exec, ssh_upload, ssh_download require session_id from ssh_connect. Tool descriptions do not state this prerequisite or guide the dependency chain.
Parameter constraints missing: Timeout parameters lack min/max bounds (risk of invalid values). Port numbers lack range constraints. File paths lack validation rules (must exist, readable, writable). LLM can pass absurd values.
Idempotency not documented: Tools like ssh_upload, ssh_key_manage 'add' lack clarity on idempotent behavior. ssh_upload, overwrite or error if exists? ssh_key_manage add, add same key twice? Agents need to know if safe to retry.
Security implications not addressed: ssh_connect accepts passwords as parameters (risk of logging credentials). No mention of secret injection, key-based auth preference, or secure secret storage. Tool descriptions do not acknowledge security stance.
Error handling guidance missing: No documented error messages or recovery paths. E.g., ssh_exec on timeout, should LLM retry? Increase timeout? What error is returned? ssh_tunnel on 'port already in use', actionable guidance?
Naming ambiguity: 'ssh_config' is unclear (load? read? apply?). 'ssh_key_manage' is a compound verb+noun (should be more specific like 'manage_ssh_keys' or split into 'add_ssh_key', 'remove_ssh_key', 'list_ssh_keys'). ssh_tunnel is generic (create_tunnel, start_tunnel more explicit).