Loading ToolBench
MCP server that provides semantic search over an indexed corpus of documents using embeddings and similarity search
Single tool with Turkish description and minimal English documentation. Schema is present with proper Zod validation (query: string min 2, limit: number 1-10 optional), but description lacks English clarity and actionable guidance. No parameter descriptions visible in schema. Output is JSON-stringified text rather than structured response. No error handling guidance, recovery paths, or pagination despite accepting a limit parameter. Tool naming is clear (search_corpus) but descriptions are in Turkish, creating accessibility issues for English-speaking LLM agents.
İndekslenmiş dokümanlarda arama yapar ve kaynak metinleri sıralı şekilde döndürür.
Tool description is in Turkish ('İndekslenmiş dokümanlarda arama yapar...'), not English. LLM agents trained on English will not understand when or why to call this tool.
Input parameters lack descriptions. 'query' and 'limit' have no explanation of expected format, constraints, or use cases. Rubric requires every parameter to have a description.
Output is returned as JSON-stringified text in a single 'text' content block, not as structured fields. LLMs cannot reliably parse and extract individual results for downstream tool calls or chaining.
No error handling or recovery guidance. If search fails, returns no error message or suggestion for next steps. LLM cannot self-correct or retry intelligently.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 44 | 2026-07-28+ | v2 |
Tool accepts 'limit' parameter but no pagination support (no offset, cursor, or total count). Returning up to 10 results without indicating if more exist prevents proper pagination.