Loading ToolBench
MCP server for Google Vertex AI search with Google Search grounding
The server defines a single tool 'vertex_search' with a clear action verb, reasonable description (127 chars), and explicit input schema. However, there are significant gaps: the output schema is completely undocumented (no field descriptions, no structure), parameters lack format/constraint details despite accepting user-defined values, and error handling is not described. The tool itself is straightforward (search with optional filters), but the implementation does not meet A+ production standards for output documentation and parameter precision.
Search the web using Google Vertex AI (Gemini) with Google Search grounding. Returns an AI-synthesized answer with inline citations.
Output schema is completely undocumented. Tool description says 'Returns an AI-synthesized answer with inline citations' but does not specify the response structure, field names, or data types. LLMs cannot reliably parse unstructured output or plan downstream actions.
Parameter 'query' lacks format/constraint details. Description says 'minLength:1' in the schema but does not explain the expected format, maximum length, or character restrictions in the description text itself. LLMs do not reliably read JSON Schema constraint fields; they rely on description text.
Parameters 'site_filter' and 'time_range' lack actionable constraint documentation. Descriptions are generic ('Optional domain to restrict search to'); no guidance on format (e.g., 'use FQDN format like stackoverflow.com, not http://stackoverflow.com'). No examples of valid time_range values ('past 24 hours' vs 'last 24h' vs 'P1D').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 53 | - | v1 |
No error handling documentation. The tool description does not explain what happens on rate limits, invalid queries, authentication failures, or network timeouts. No guidance for LLM recovery ('If you get a rate limit, wait and retry' or 'If authentication fails, check API key configuration').
Tool description is borderline short (127 chars, within the 10 - 1024 range but below the 194-char baseline for A+ tools). It lacks explicit guidance on WHEN to use this tool vs. alternatives, and what the LLM should expect from the result (synthesis quality, citation reliability, speed/latency).