MCP server that makes any website usable by AI agents — filters 90% of DOM noise, returns structured element inventory with stable locators
LeftGlove provides 7 browser automation tools with basic descriptions but critical gaps in schema documentation and parameter clarity. Tool names are action-oriented (observe, click, fill, navigate, screenshot, element-screenshots, status) but parameter schemas are incomplete or inferred from scattered source files. Most tools lack formal input schema definitions visible in the code. Descriptions are present but generic (10-80 chars), below the 50-200 char baseline for LLM optimization. No output schemas documented. Error handling and recovery guidance absent. The observe tool is well-named but lacks schema detail; click/fill accept multiple input modes (index, coordinates, selector) without clear mutual exclusivity documentation. No tool declares permissions, audit trails, or idempotency guarantees.
Click an element by index, coordinates, or CSS selector
Take per-element screenshots by index
Fill an input element with text by index, coordinates, or CSS selector
Navigate to a URL
Inject sieve.js into the current page, return structured element inventory with stable locators
Take a PNG screenshot of the current page
Get current browser state and page information
No visible input schemas for observe, screenshot, status tools. Cannot verify parameter types or constraints.
click and fill accept three mutually exclusive input modes (index, coordinates, selector) but lack explicit mutual-exclusivity documentation or enum constraints. LLMs will pass multiple modes simultaneously, causing ambiguous behavior.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., element indices from observe, success/failure from click). Breaks tool composition.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Descriptions are 35-50 characters, below the 50-200 char baseline. Too brief to guide LLM selection. E.g., 'Take a PNG screenshot of the current page' lacks context on when to use element-screenshots vs screenshot.
No error handling or recovery guidance. Tools do not document what happens on failure (e.g., click on non-existent element, navigate to invalid URL). LLMs cannot self-correct.