MCP server for browser automation using cloudbrowser
8 tools with basic schemas and minimal descriptions. All tools have input schemas with required fields properly declared, but descriptions are generic (10-50 chars) and lack context for LLM selection. No parameter descriptions beyond selector/value/script. No output schemas documented. No error handling guidance. Tool names follow verb_noun pattern (navigate, evaluate, click) but lack specificity about what state changes occur. Missing idempotency, retry guidance, and recovery instructions.
Click an element on the page
Evaluate JavaScript in the browser
Fill out an input field
Retrieve the current URL of the browser page
Extract all text content from the current page
Hover an element on the page
Navigate to a URL
No output schemas documented. LLMs cannot plan downstream calls or extract return values. E.g., yunbrowser_screenshot returns what? Base64 image? URL? File path?
Descriptions are 10-50 characters, below the 194-char baseline. Generic text like 'Navigate to a URL' lacks context for when to use this tool vs alternatives, prerequisites, or side effects.
Parameter descriptions missing for most inputs. 'selector' appears in 4 tools but is never described (CSS selector format, required? optional? error if not found?). 'script' in yunbrowser_evaluate has no description.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
Take a screenshot of the current page or a specific element
No error handling guidance. If yunbrowser_click fails (element not found, page crashed, timeout), what should the LLM do? Retry? Call a different tool? No recovery path documented.
No idempotency or confirmation for destructive operations. yunbrowser_evaluate can execute arbitrary JavaScript (side effects unknown). yunbrowser_navigate changes page state. No dry-run, confirmation, or undo mechanism.