Loading ToolBench
Integrate RAG into AI Agents via MCP Server. Supports multiple Vector database technologies.
Both tools exhibit critical definition quality gaps. Tool names lack action verbs (vector_collection_management, vector_search are nouns, not verb_noun patterns). Descriptions are verbose and list actions as bullet points rather than explaining WHAT the tool does and WHEN to use it. Parameters lack individual descriptions, only bare names like 'db_type', 'host', 'port' with minimal context. No output schemas documented. The 'action' parameter is a free-form string enum that should be split into separate, focused tools (create_collection, add_documents, delete_collection, list_collections, semantic_search, lexical_search, hybrid_search). Error handling and recovery guidance are absent.
Manage collection management operations. Actions: - 'create_collection': Creates a new collection or retrieves an existing one in the vector database. - 'add_documents': Adds documents to an existing collection in the vector database. - 'delete_collection': Deletes a collection from the vector database. - 'list_collections': Lists all collections in the vector database.
Manage search operations. Actions: - 'semantic_search': Retrieves and gathers related knowledge from the vector database instance using the question variable. - 'lexical_search': This is a lexical or term based search that retrieves and gathers related knowledge from the database instance using the question variable via BM25. - 'search': Performs a hybrid search combining semantic (vector) and lexical (BM25) methods.
Tool names are nouns, not action verbs. 'vector_collection_management' and 'vector_search' do not follow verb_noun convention. Should be split into: create_collection, add_documents, delete_collection, list_collections, semantic_search, lexical_search, hybrid_search.
No output schemas documented. LLMs cannot plan downstream calls or extract required fields (e.g., collection_id, search_results structure). Every tool must document return type and fields.
Parameters lack individual descriptions. 'db_type', 'host', 'port', 'username', 'password' have only bare names, no guidance on format, constraints, or when to use each. Descriptions must explain what each parameter controls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | <=2025-11-25 | v2 |
Credentials (username, password) exposed as tool parameters. Must use server-side secret injection via environment variables or vault. Agent traces log all parameters, secrets leak into logs and prompt history.
No error handling or recovery guidance. Tools do not indicate what to do on failure (retryable? user-fixable? fatal?). No actionable error messages. Agents cannot self-correct.