MCP Server for macOS desktop control via screencapture, cliclick, and osascript
14 tools with complete input schemas and descriptions. Naming is action-verb based (screenshot, click, type, key, scroll, drag, focusApp, getMousePosition, getActiveApp, getScreenSize, listApps, getWindowInfo, getTile). Descriptions are present and substantive (100-250 chars typical), explaining WHAT each tool does and WHEN to use it. Parameters are well-typed with enums for constrained inputs (compression presets, mouse buttons, modifier keys, image formats). However, output schemas are not documented, LLMs cannot see what fields to expect from responses. Error handling is minimal; no recovery guidance or actionable error messages visible in code. Security is reasonable (no secrets in params, read-only vs write risk classification present), but no permission gates or audit logging. Composition is strong, tools are single-responsibility and chainable (e.g., screenshot → getTile, click → focusApp). Baseline: 549 production tools average 18-char names (these average 12 chars, good); 194-char descriptions (these average 160 chars, slightly short but acceptable); 100% of A+ tools have param descriptions (this server achieves ~95%).
Click at screen coordinates. Supports modifiers (cmd, alt, ctrl, shift) and multiple clicks. Foreground mode: brings target app to focus.
Drag from one position to another. Supports modifiers and custom duration.
Bring an application to the foreground and activate it.
Get the name of the currently active (frontmost) application.
Get the current mouse cursor position on screen.
Get the primary screen resolution (width and height in pixels).
Retrieve a specific tile from a previous tiled screenshot. Tiles expire after 2 minutes.
Get window information (bounds, PID, name) for a target application.
Output schemas not documented. LLMs cannot see what fields screenshot, getTile, listApps, getWindowInfo, and other tools return. This forces LLMs to guess field names and types, increasing errors and wasting tokens on exploratory calls.
No error handling guidance. Code does not show recovery hints (e.g., 'If app not found, try listApps()') or error categorization (retryable vs user-fixable). Agents cannot self-correct on failures.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). While risk classification is present in metadata, MCP tool annotations are missing, limiting client-side safety checks and UI hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Press keyboard keys. Supports modifiers (cmd, alt, ctrl, shift) and key combinations.
List all running applications on the system.
Move mouse to screen coordinates. Supports absolute or relative positioning.
Capture the entire screen or a specific region. Returns image data (PNG/JPEG/WebP) with optional compression and tiling.
Scroll at screen coordinates. Supports horizontal and vertical scrolling.
Type text at the current cursor position. Supports modifiers for special keys.
Foreground mode (app + windowTitle + relativeX/Y) parameters are optional and undocumented in terms of mutual exclusivity. LLMs may pass both absolute and relative coords, causing ambiguous behavior.
getTile description does not explain tile expiration (2 minutes) or what happens if a tile is requested after expiry. LLMs may retry expired tiles without understanding the root cause.