Four tools with complete schemas and descriptions. Naming follows verb_noun pattern (ssh_list_hosts, ssh_exec, ssh_upload, ssh_download). Descriptions are clear and contextual (avg ~180 chars), explaining WHAT, WHEN, and prerequisites. All parameters have type definitions and descriptions. However, output schemas are not formally documented, responses are returned as text blobs rather than structured objects. Error handling is minimal (generic 'Error: {message}' responses). No input validation guidance or recovery hints. Missing tool annotations (readOnlyHint/destructiveHint). Parameter descriptions lack format constraints (e.g., no guidance on valid host names, command syntax, or file size limits).

• Document output schemas for all tools. ssh_list_hosts should return {hosts: [{name: string, address: string, user: string}]}. ssh_exec should return {host: string, exit_code: number, stdout: string, stderr: string, signal?: string}. Use structured response objects, not text blobs.
• Add tool annotations via server.tool() metadata: ssh_list_hosts and ssh_download with readOnlyHint=true; ssh_exec and ssh_upload with destructiveHint=true. This signals to agents which operations are safe to retry.
• Enhance error messages with recovery guidance. E.g., 'Host "prod-server" not found. Call ssh_list_hosts to see available hosts.' or 'Command rejected by user: {reason}. You may retry with a modified command or ask the user for approval.'
• Add format constraints to parameter descriptions. E.g., host: 'Name of a configured server (call ssh_list_hosts to discover valid names)'. command: 'Shell command to execute (e.g., ls -la /var/log)'. remote_path: 'Absolute path on the server (must start with /).'
• Clarify mutually exclusive parameters in ssh_upload: 'Provide either inline content OR local_path, not both. If both are provided, content takes precedence.'