A browser automation MCP server for AI models like Claude and Gemini 2.5
Critical gaps across all dimensions. Tool descriptions are present but generic (avg ~40 chars, well below 194-char baseline). Input schemas visible but lack parameter descriptions, most params have no explanation of what they control or valid ranges. No output schemas documented. sessionId parameter appears in all tools but is never explained. Error handling absent. No tool annotations (readOnlyHint/destructiveHint). Naming is reasonable (verb_noun pattern) but parameter naming inconsistent (selector vs fields). This server would not pass code review for production use.
Analyze webpage content using AI
Browse a webpage and extract its content
Click on an element on the webpage
Extract content from the current webpage
Fill out form fields on a webpage
Search the web for information
Take a screenshot of the current webpage
No parameter descriptions. All 7 tools expose parameters (url, query, selector, fields, etc.) with type but zero explanation of what each controls, valid ranges, or format constraints. LLMs cannot infer intent from bare parameter names.
sessionId parameter unexplained across all tools. Appears in every tool but no description of what it is, how to obtain it, or whether it's required. Forces LLM to guess.
No output schemas documented. Tools return data but LLMs have no specification of response structure, field types, or what to expect. Breaks downstream tool chaining.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | <=2025-11-25 | v2 |
No error handling guidance. No recovery hints, error classification, or actionable messages. If a tool fails, LLM has no direction on retry, fallback, or next steps.
No tool annotations. Destructive tools (click_element, fill_form) lack destructiveHint. Read-only tools lack readOnlyHint. LLMs cannot distinguish safe from risky operations.