MCP server for automated test generation using Playwright browser automation and Claude AI. Explores web pages, extracts interactive elements, and generates Playwright tests with verified selectors.
Server has 5 tools with basic schemas and descriptions in Spanish. All tools have input schemas with type definitions and descriptions present. However, descriptions are minimal (10-50 chars), lack context on WHEN to use tools, and omit error recovery guidance. Parameters lack constraints (enums, ranges, formats). No output schemas documented. Error handling returns generic text responses without actionable guidance. Tools lack idempotency hints and state-modification warnings. No tool annotations (readOnlyHint/destructiveHint). Naming is verb-based but descriptions are too terse to guide LLM selection effectively.
Cierra el navegador
Encuentra elementos en la página usando un selector CSS
Obtiene el contenido HTML de la página actual
Navega a una URL específica
Toma una captura de pantalla de la página actual
Descriptions are too brief (10-50 chars) and lack context. 'Navega a una URL específica' does not explain WHEN to use navigate vs other tools, what happens on failure, or prerequisites. LLMs cannot reliably select tools without 50-200 char descriptions explaining purpose, use case, and dependencies.
No output schemas documented. Callers cannot predict response structure. Tools return TextContent with unstructured text (e.g., 'HTML Content:\n{content}'), forcing LLMs to parse free-form strings. Structured JSON responses with typed fields enable reliable downstream tool chaining.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). screenshot and close_browser modify state but lack destructiveHint. LLMs cannot distinguish safe read-only tools from irreversible operations, risking accidental data loss.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
Error handling returns generic text ('❌ Error navigando: {str(e)}') without recovery guidance. LLMs receive no actionable next steps. Should return: error category (retryable/user-fixable/fatal), the invalid value, constraint violated, and suggested recovery action.
Parameters lack constraints. 'url' in navigate has no format validation (http/https only?). 'path' in screenshot has no directory restrictions (path traversal risk?). 'selector' in find_elements has no length limit. Free-form strings invite invalid input and security issues.