Loading ToolBench
MCP Server for Nordstemmen documents via Qdrant vector database, providing semantic search and document retrieval capabilities
The Nordstemmen server defines 5 tools with reasonable naming conventions and complete input schemas. However, descriptions are generic and lack LLM-optimization guidance. Parameter descriptions exist but are minimal (typically 1-2 lines). No output schemas are documented. Error handling guidance is absent. Tools follow verb_noun patterns (search*, get*) which is positive, but descriptions do not explain WHEN to use each tool vs alternatives, WHY the agent should call it, or what the expected response structure contains. This is typical of C-grade community servers, functional but not optimized for agent reasoning.
Retrieve the full text content of a document identified by file hash
Retrieve a specific paper by its reference identifier, normalizing the reference format
Search documents by semantic query with optional date range filtering, combining dense and sparse vector search with RRF fusion
Search meetings by various criteria including committee, start date, and organizer
Search papers by multiple criteria including reference pattern, name, type, and date range
Missing output schemas. No documentation of return types, field names, or structures for any tool. LLMs cannot plan downstream calls or extract relevant data without knowing response shape.
Descriptions lack LLM optimization guidance. Descriptions are factual but do not explain WHEN to use each tool vs similar ones, or what prerequisites/dependencies exist. E.g., searchDocuments vs searchPapers are both search operations, the descriptions should clarify which is semantic vs metadata-based. Current descriptions (48-54 chars avg) are at the lower end of the 10-1024 char range and lack context for selection.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 56 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Parameter descriptions are minimal and lack formatting constraints. E.g., 'limit' parameters lack min/max bounds (stated as 'default: 5, max: 10' in natural language, not schema). Date parameters lack ISO 8601 format specification. 'reference_pattern' and 'committee_pattern' accept regex but do not document whether the regex syntax is ECMA262, PCRE, or other.
No error handling guidance. Tools do not document error conditions, recovery paths, or actionable error messages. E.g., if getFulltext fails (file_hash not found), what should the agent do? Call searchDocuments first? Retry? Treat as fatal?
Pagination and result limiting unclear. searchDocuments and searchPapers mention 'limit' parameters and defaults, but do not document total count or cursor-based pagination for large result sets. If a query returns 500+ papers, how does the agent paginate?
No tool interdependency documentation. searchPapers and getFulltext both take identifiers but do not explain: does getFulltext require output from searchPapers? If getPaperByReference accepts 'DS 2024/1234' or '2024/1234', what does searchPapers return, which format?