Loading ToolBench
MCP Server integrating Pinecone (RAG), Groq (LLM), and Linkup (Web Search) with PDF support
Server has 4 tools with basic schemas and descriptions, but significant gaps limit production readiness. Tool names lack clear action verbs (query, vision, add_document are acceptable but generic). Descriptions exist but are brief (10-50 chars) and lack context on when/why to use each tool. Input schemas are present with type definitions, but parameter descriptions are minimal. Output schemas are not documented. Error handling is present in code but not surfaced in tool definitions. No tool annotations (readOnlyHint, destructiveHint). The add_document tool modifies state but lacks confirmation/dry-run pattern.
Add a document to the knowledge base
Health check endpoint
Process a query with optional RAG and web search
Process an image using Groq's vision capabilities
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields. query returns {answer, provider, used_rag, used_web_search} but this is inferred from code, not declared in tool definition.
Parameter descriptions are minimal or missing context. 'query' param lacks guidance on format/length. 'use_rag' and 'use_web_search' lack explanation of when to enable each. 'prompt' in vision tool has only 'Text prompt to accompany the image', no guidance on what makes a good prompt.
add_document is a destructive/write operation but lacks confirmation pattern, dry-run option, or explicit warning in description. Tool description is only 35 chars ('Add a document to the knowledge base'). No guidance on file size limits, supported formats beyond the schema, or failure modes.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | <=2025-11-25 | v2 |
Tool descriptions are too brief (10-50 chars). Baseline for A+ tools is 194 chars. 'Process a query with optional RAG and web search' (50 chars) lacks context on when to use RAG vs web search, what RAG requires, or expected latency.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). add_document should be marked destructive; query and vision should be marked readOnly. This prevents agents from reasoning about side effects and retry safety.