Loading ToolBench
MCP server exposing Lumina RAG tools for querying knowledge base and listing documents
Two tools with minimal schema documentation and generic descriptions. list_documents has empty input schema ({}), violating the requirement that every parameter needs a description. query_knowledge_base has three parameters with type strings and descriptions present, but lacks output schema documentation. Neither tool description exceeds 100 characters, and descriptions lack actionable context about when to use each tool or what data structures are returned. No error handling guidance, no pagination support despite query_knowledge_base likely returning variable-length results, and no indication of result limits. Tool naming follows verb_noun convention (positive), but descriptions are too generic to guide LLM selection.
List all uploaded documents in the Lumina RAG knowledge base.
Query the Lumina RAG knowledge base for text segments, tables, and visual captions.
list_documents has empty input schema ({}). Per HARD SCORING RULE, schema score must be 0. Tool provides no parameters but also no documentation of what it returns or how results are structured.
query_knowledge_base lacks output schema documentation. LLMs cannot plan downstream calls or extract relevant fields without knowing the response structure (e.g., does it return text chunks, table objects, captions separately, or combined?).
No pagination support visible. query_knowledge_base likely returns variable-length results but lacks limit, offset, page, or cursor parameters. Large result sets will blow context windows.
Descriptions are generic and under 100 characters. 'List all uploaded documents' and 'Query the Lumina RAG knowledge base' do not explain WHEN to use each tool, what distinguishes them, or what data structures are returned. LLMs cannot reliably select between them.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 42 | 2026-07-28+ | v2 |
No error handling guidance. If query_knowledge_base returns no results, or if session_id is invalid, the tool provides no recovery hints. LLMs cannot self-correct.