Automation MCP, to automate inputs give models access to your computer with detailed control.
This automation server provides 10 tools with reasonable naming and parameter coverage. However, it suffers from significant definition quality gaps: tool descriptions lack depth and context for LLM decision-making; output schemas are completely undocumented; error handling is minimal and non-actionable; and critical parameters lack type constraints. Most tools fall into the 'fair' range (50-65) due to present-but-incomplete schemas and generic descriptions. The server's purpose (desktop automation) is clear, but individual tool utility descriptions do not explain WHEN to use each tool vs. alternatives, or what errors mean and how to recover. No tool describes its output structure, forcing LLMs to guess at response format. Baseline expectation: avg description length 194 chars; this server averages ~100 chars and omits output documentation entirely.
Press or release specific keys for advanced key combinations.
Press or release a mouse button without clicking.
Simulate a mouse click at the given screen coordinates.
Simulate a mouse double-click at the given screen coordinates.
Drag the mouse from current position to target coordinates.
Get the current mouse cursor position.
Move the mouse to specific coordinates.
No output schemas documented for any tool. LLMs have no way to know what fields to expect or plan downstream tool calls. Pattern:response-shaper requires documented return types; 100% of A+ tools have them.
Tool descriptions are too generic and lack context. 'Simulate a mouse click' (50 chars) does not explain WHEN to use mouseClick vs mouseDoubleClick, or that both move the cursor. Baseline is 194 chars avg; descriptions should explain use case and relationship to similar tools.
Parameter 'type' in screenshot tool is ambiguous. Enum values are ['full', 'region', 'window'] but description does not explain what 'type' means or why an LLM would choose one over the other. Requires context-dependent reasoning without guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 21 | - | v1 |
Scroll the mouse wheel in a specified direction.
Capture a screenshot (full screen, region, or window).
Simulate typing text or pressing key combinations. Provide either 'text' to type literal text, or 'keys' as comma-separated key names for key combinations.
screenshot tool parameters 'region' and 'windowHandle' are declared as required conditionally ('required if type=region') but JSON Schema does not enforce this dependency. Description warns of it, but LLM must infer the relationship. Should use conditional schema or explicit documentation.
No error handling guidance. Tools throw errors on nutjs unavailability or invalid input but do not indicate whether errors are retryable, user-fixable, or fatal. Pattern:recovery-guide requires actionable error messages; this server omits them.
'type' tool parameter naming is ambiguous. Tool accepts 'text' OR 'keys' but both are optional, forcing LLM to reason about mutually exclusive behavior. Should document which combinations are valid or split into separate tools (type_text, press_keys).
mouseGetPosition returns unstructured string 'Current mouse position: (x, y)' instead of structured JSON {x: number, y: number}. Pattern:response-shaper requires structured objects; free-text responses waste tokens and force LLM parsing.
All tools return plain-text success strings instead of structured results. 'Mouse click at (100, 200) completed' provides no machine-readable confirmation or result metadata. LLMs cannot reliably extract success state or compose tool chains.
Numeric parameters (x, y, amount) lack min/max bounds. An LLM could pass negative coordinates or absurdly large scroll amounts without validation. Rubric requires: 'Specify minimum and maximum for numeric parameters.'