An MCP Server for RAG Web Browser Actor
Single tool 'search' has well-structured schema with proper types, enums, and descriptions. Tool name is action-oriented (verb_noun pattern). Description is clear and actionable (145 chars, within 10-1024 baseline). All 5 parameters have type definitions and descriptions. However, output schema is not documented, callers cannot predict response structure. Error handling is minimal (generic 'Failed to call RAG Web Browser' with no recovery guidance). No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Parameter descriptions could be more concise and LLM-optimized.
Search phrase or a URL at Google and return crawled web pages as text or Markdown. Prefer HTTP raw client for speed and browser-playwright for reliability.
Output schema not documented. Callers cannot predict response structure or plan downstream tool calls. LLMs must guess what fields are returned.
Error messages lack recovery guidance. 'Failed to call RAG Web Browser: 401 Unauthorized' tells the LLM nothing actionable. Should suggest 'Check APIFY_TOKEN is set and valid.'
No tool annotations. 'search' is read-only and idempotent, but these hints are missing. LLMs cannot determine if the tool is safe to retry or has side effects.
Parameter descriptions are verbose and include implementation details ('Prefer HTTP raw client for speed'). Should focus on what the LLM needs to know to invoke correctly.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
No pagination or result limits documented. If RAG Web Browser returns large responses, context window could be exhausted. Should cap results and document limit in description.