Server defines 25 tools with complete JSON schemas and French descriptions. Naming follows verb_noun convention (get_cursor_position, move_mouse, screenshot). Schemas are well-formed with proper types, enums, and required fields. However, descriptions are inconsistent in depth, some lack context on when to use the tool or what it returns. Parameter descriptions are present but terse (avg ~40 chars vs baseline 72). No output schemas documented. Error handling exists but lacks recovery guidance. Security concerns: run_command exposes shell execution without clear permission gates or sanitization warnings.

• Document output schemas for all tools. E.g., get_cursor_position returns {x: integer, y: integer}; list_windows returns [{hwnd: integer, title: string, x: integer, y: integer, width: integer, height: integer}]. This enables LLM planning and chaining.
• Add tool annotations: mark screenshot, get_cursor_position, list_windows as readOnlyHint=true; mark move_mouse, click, run_command as destructiveHint=true; mark idempotent operations accordingly. Helps LLMs reason about retry safety.
• Expand run_command description to warn about injection risks and require explicit user confirmation for destructive operations. Add a dry_run parameter or separate preview_command tool.
• Enhance parameter descriptions with ranges and formats. E.g., 'x: integer (0 to screen_width, absolute pixel coordinate)' instead of 'Coordonnee X absolue'. Reference get_screen_size for bounds.
• Add recovery hints to error messages. E.g., 'Enigo initialization failed: ensure display server is running. Try calling get_screen_size() to verify screen access.' This guides LLM retry logic.