Loading ToolBench
Custom MCP server exposing Pinecone vector search as an agentic-search tool
Single tool 'agentic-search' has a clear verb-noun name and reasonable description (127 chars, within 10-1024 range). Input schema is properly defined with Zod types and descriptions for both parameters. However, output schema is undocumented, the tool returns text content but LLMs cannot see the structured response format. Error handling returns isError flag but lacks recovery guidance. No tool annotations (readOnlyHint, etc.) despite being a read-only operation. Parameter descriptions are adequate but could be more prescriptive about constraints (e.g., topK range, query length limits).
Search the GenAI knowledge base covering AI Agents, RAG, MCP, and Prompt Engineering. Returns relevant excerpts from ArXiv research papers.
Output schema undocumented. Tool returns {content: [{type: 'text', text: string}], isError?: boolean} but LLMs cannot see this structure in the tool definition. Agents cannot reliably extract or chain results.
No tool annotations. Tool is read-only and idempotent but lacks readOnlyHint annotation. Agents cannot infer safety properties without explicit hints.
Parameter constraints missing. topK parameter has no min/max bounds (e.g., 1-100). Query parameter has no length limit. Unbounded inputs risk API abuse or timeout.
Error messages lack recovery guidance. 'Search failed: <message>' tells LLM nothing about whether to retry, ask user, or abandon. No error classification (retryable vs fatal).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
No pagination support. Tool accepts topK but does not return cursor/offset for continuation. Large result sets cannot be iterated; agent cannot fetch next page of results.