SSH MCP server demonstrates solid definition quality with consistent naming conventions, explicit schemas, and comprehensive parameter descriptions. All 11 tools are properly defined with input schemas and descriptions. Tool names follow verb_noun patterns (read-command, run-command, sftp-upload, etc.). Schemas are well-structured with proper types and constraints. However, output schemas are not explicitly documented in the visible source, descriptions are mostly functional but lack LLM-optimized guidance on when/why to use each tool, and error handling recovery guidance is minimal. Risk annotations (READ_ONLY, WRITE, DESTRUCTIVE) demonstrate security awareness but lack permission gate documentation. The server shows production-ready implementation patterns but misses some polish in guidance and composition.

• Document output schemas explicitly for all tools. Example: 'Returns {success: boolean, output: string, exitCode: number, stderr?: string}' in tool descriptions or via explicit schema fields if SDK supports.
• Enhance descriptions with WHEN/WHY guidance. Example for read-command: 'Use for safe, read-only queries against the remote system. Calls are restricted to a security allowlist. Returns command output and exit status.'
• Add recovery hints to command execution tools: 'If the command fails, check exit code and stderr. Common errors: command not found (exitCode 127), permission denied (exitCode 126). Retry with different parameters or verify remote state with read-command.'
• Explicitly state permission/scope requirements in tool descriptions. Example for privileged-command: 'Requires sudo_write scope. User must be in sudoers on the remote host.'
• Add dependency notes for session tools: 'Call open-session first to create a session, then use run-command with session=<name> for multi-step workflows, or close-session when done.'
• Document idempotency guarantees. Example: 'read-command is idempotent, same command always returns same output. run-command may have side effects, use cautiously in loops.'
• Consider adding a 'profile' resolution tool or describe default profile behavior clearly so agents understand how profiles are selected when omitted.