This server has 12 tools with basic definitions but significant quality gaps. All tools have names and descriptions, but descriptions are minimal (10-60 chars, well below the 194-char baseline). Most critically, 10 of 12 tools have EMPTY input schemas (just `{}`) with no parameter documentation. The 2 tools with parameters (claude_send_message, claude_send_long_message, claude_shortcut) have malformed schema definitions that nest JSON Schema incorrectly. No output schemas are documented. Error handling is minimal, most tools return plain text with no recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk distinctions. The server lacks idempotency guarantees, rate limiting, and permission gates for destructive operations (quit, launch, send_message).

• Fix JSON Schema syntax in claude_send_message, claude_send_long_message, claude_shortcut. Use proper JSON Schema: {"type": "object", "properties": {...}, "required": [...]} instead of nesting as object values.
• Expand all tool descriptions to 100-200 characters. Include WHAT the tool does, WHEN to use it, and what it returns. Example: 'Send a message to Claude Desktop by typing and pressing Enter. Use for short messages; use claude_send_long_message for complex text with special characters.'
• Add parameter descriptions to all tools. For claude_send_message: 'message (string, required): The text to send to Claude. Will be typed character-by-character and Enter pressed.' For claude_shortcut: 'key (string, required): Single key to press (e.g., "n", "c", "v"). modifiers (array, optional): Modifier keys from ["command", "shift", "option", "control"].'
• Add tool annotations: mark claude_is_running, claude_get_window_info, claude_get_ui_elements, claude_copy_conversation, claude_screenshot as readOnlyHint=true. Mark claude_launch, claude_quit, claude_activate, claude_new_conversation, claude_send_message, claude_send_long_message, claude_shortcut as destructiveHint=true.
• Document output schemas for all tools. Example: claude_is_running returns {"status": "running" | "not_running"}. claude_get_window_info returns {"x": number, "y": number, "width": number, "height": number}. Add error recovery guidance: 'If Claude is not running, call claude_launch first.'