Server has 3 tools with basic schemas and descriptions, but significant quality gaps. Tool naming lacks consistency (run_command vs add_numbers vs convert_image_to_greyscale, mixed verb_noun patterns). Descriptions are present but generic (10-50 chars for most params). No output schemas documented. run_command is destructive but lacks confirmation/dry-run pattern. Parameter descriptions are minimal or absent. No error handling guidance. Schemas are present but incomplete, missing type info for some fields, no validation constraints, no pagination support.

• Add output schema documentation for each tool. E.g., add_numbers returns {type: 'number', description: 'The sum of a and b'}. convert_image_to_greyscale returns {type: 'string', format: 'data-uri', description: 'Base64-encoded greyscale JPEG as data URI'}.
• Implement confirmation pattern for run_command: add optional 'dry_run' boolean param (default false) that echoes the command without executing. Require explicit confirmation for destructive commands (rm, dd, etc.).
• Enhance parameter descriptions with format, constraints, and examples. E.g., command: 'Shell command to execute (no pipes, redirects, or subshells; max 500 chars). Examples: ls -la, pwd, whoami. Avoid: rm -rf, dd, sudo.'
• Add error handling and recovery guidance. E.g., run_command should return {success: bool, output: string, error_type: 'timeout|permission|not_found|syntax_error', suggestion: string}. convert_image_to_greyscale should catch HTTP errors and return actionable messages like 'Image URL unreachable. Check URL and try again.'
• Sanitize run_command input: validate against a whitelist of safe commands or use shlex.quote() to escape arguments. Document which commands are blocked (rm, dd, sudo, etc.) in the description.