An MCP server that provides control over screen, mouse, and keyboard using nut.js
Server provides 23 tools with consistent naming (verb_noun pattern) and complete input schemas using Zod. All tools have descriptions and parameters are typed with enums where appropriate. However, descriptions are uniformly terse (10-50 chars), falling below the 50-200 char baseline for LLM-optimized descriptions. Output schemas are not documented, responses are hardcoded text wrapping rather than structured objects. Error handling returns stack traces but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification. Missing parameter descriptions for some optional fields.
Capture a specific region of the screen
Capture the entire screen as an image
Click the mouse at a specific position
Click the mouse at the current position
Copy selected text to clipboard using keyboard shortcut then return the clipboard content
Double-click the mouse at the current position
Double-click the mouse at a specific position
Descriptions uniformly terse (10-50 chars) vs. baseline 50-200 chars. LLMs cannot determine when/why to select tools. E.g., 'Capture the entire screen as an image' lacks context on output format, use cases, or when to prefer capture_region.
Output schemas not documented. All tools return hardcoded text responses ('Screen captured successfully...') rather than structured objects. LLMs cannot parse results or chain tools. No field definitions for image data, coordinates, or status.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Drag the mouse from current position to target position
Drag the mouse from start position to end position
Get text from the clipboard
Get the current position of the mouse cursor
Get the dimensions of the screen
Hold down a key
Move the mouse cursor to a specific position
Paste text from clipboard using keyboard shortcut
Set text to clipboard then paste it at current cursor position
Press a keyboard key
Press a keyboard shortcut (combination of keys)
Release a key that's being held down
Scroll the mouse wheel
Set text to the clipboard
Type text at the current cursor position
Type text with delays between keystrokes
No tool annotations despite clear risk stratification. 13 WRITE tools (click_mouse, type_text, press_key, etc.) and 10 READ_ONLY tools lack readOnlyHint/destructiveHint. Agents cannot reason about side effects or retry safety.
Error handling returns stack traces and raw error messages without recovery guidance. E.g., 'Error capturing screen: ...' does not tell LLM whether to retry, check permissions, or call a different tool. No error classification (retryable vs. fatal).
STDIO transport only. Server is not remotely accessible and cannot be used by hosted MCP clients. Limits deployment to local/desktop environments.