Loading ToolBench
Mingle with all AI minds in one place - Electron desktop application with MCP server integration for RAG operations and multi-LLM support
Mingly's RAG server exposes 13 tools with significant quality gaps. Tool definitions exist but lack consistency: some have detailed schemas (search_documents, index_document), others are minimal stubs (list_collections, get_stats). Descriptions range from 50 - 200 chars but many lack actionable context. Critical issue: 6 tools appear duplicated with inconsistent naming (search_documents vs search, index_document appears twice with different param names: file_path vs filepath). Parameters mostly have types and descriptions, but no enum constraints on collection names or validation guidance. Output schemas are undocumented, LLMs cannot predict response structure. Error handling is absent; no recovery guidance. Security concern: file_path parameters accept absolute paths with no sanitization hints, risking path traversal. Composition is poor: overlapping tools (search vs search_documents, list_collections appears twice) force LLM disambiguation overhead.
Create new collection
Delete collection
Delete an entire collection and all its documents. USE WITH CAUTION!
Get detailed statistics for a collection: document count, vector count, indexed files.
Get formatted context string ready for LLM system prompt injection. Searches for relevant documents and returns pre-formatted context.
Get collection statistics
Index all supported documents in a directory (recursively). Useful for bulk-importing document libraries.
Index a single document (PDF, DOCX, PPTX, MD, TXT)
Duplicate tool definitions with inconsistent naming: index_document appears twice (file_path vs filepath params), search_documents vs search, list_collections appears twice. LLMs cannot disambiguate; forces unnecessary reasoning overhead and risks wrong tool selection.
No output schemas documented. LLMs cannot predict response structure (e.g., does search_documents return {results: [{text, score, source}]} or {chunks: [...]}?). Blocks downstream tool chaining and forces LLMs to guess field names.
File path parameters (file_path, filepath, directory_path) accept absolute paths with no sanitization guidance or path traversal warnings. Risk of agents passing ../../../etc/passwd or similar. No validation hints in descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Index a document file into the RAG system. Supported formats: PDF, TXT, MD, DOCX, PPTX, HTML. The document will be chunked and embedded for semantic search.
List all collections
List all available document collections in the RAG system. Shows collection names and document counts.
Semantic search in collection
Search for relevant documents using semantic similarity. Returns the most relevant text chunks from the specified collection. Use this tool to find information in your knowledge base.
Minimal descriptions on 5 tools (list_collections, search, get_stats, delete_collection variants): 30 - 40 chars, under the 50-char minimum for LLM clarity. Agents cannot determine when to call these vs similar tools.
No enum constraints on collection parameter. Agents can pass arbitrary collection names; no guidance on valid options. Should return available collections or validate against a whitelist.