Loading ToolBench
A secure, fragment-based memory MCP server with multi-tenant support, LLM-powered reflection, and advanced memory consolidation.
Weasley DeepMind has 20 tools with basic schemas and descriptions, but significant quality gaps prevent production readiness. Tool names lack consistent verb-noun structure (e.g., 'remember' vs 'recall' vs 'forget', inconsistent action verbs; 'batch_remember' and 'batch_status' mix naming conventions). Descriptions are present but generic (avg ~60 chars, well below the 194-char baseline for A+ tools). Parameter descriptions exist but are minimal, e.g., 'Fragment ID to delete' lacks format/constraint guidance. Critical tools like 'forget' (destructive) and 'memory_consolidate' (destructive, requires master key) lack confirmation/dry-run patterns. Output schemas are not documented in the source. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. Tool composition is reasonable (single responsibility), but chaining IDs are not guaranteed in responses. Security: master key gating exists for sensitive tools, but no audit trail, rate limiting, or input sanitization visible.
Update an existing memory fragment.
지정된 단계의 업데이트를 실행합니다. 사용자 동의 후 AI가 호출. master key 전용.
Store multiple memory fragments in a single batch operation.
Check the status of a batch remember operation.
현재 버전과 최신 GitHub 태그를 비교하여 업데이트 가용 여부를 확인합니다. master key 전용.
Load contextual memory fragments for the current session.
Delete a memory fragment by ID or topic.
Destructive tools (forget, memory_consolidate) lack confirmation/dry-run patterns. No mechanism to prevent accidental data loss.
Output schemas not documented. LLMs cannot infer what fields to expect from tool responses, breaking downstream chaining and context planning.
No error handling guidance. Tools return errors without recovery hints (e.g., 'Fragment not found' vs 'Fragment not found. Try search_traces() with a keyword.').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | <=2025-11-25 | v2 |
Retrieve the edit history of a memory fragment.
Retrieve skill guide documentation for memory system usage.
Explore the semantic graph of memory fragments starting from a given fragment.
Create a semantic link between two memory fragments.
Execute memory consolidation to merge redundant fragments and optimize storage.
Retrieve memory statistics and analytics.
Retrieve memory fragments by keywords and/or topic with semantic ranking.
Reconstruct historical context for a case or entity from memory fragments.
Trigger LLM-powered reflection on session memories to extract insights.
Store a memory fragment with topic, type, content, and optional keywords.
Search memory traces by keyword and optional case ID.
Rotate the current session to start a new memory context.
Record feedback on tool relevance and sufficiency for memory optimization.
Inconsistent naming conventions. Tools use bare nouns (remember, recall, forget) instead of verb_noun pattern (store_memory, retrieve_memory, delete_memory). LLMs struggle to infer intent from names alone.
Parameter descriptions are minimal (avg ~30 chars vs 72-char baseline). Missing format constraints, ranges, and dependency hints. E.g., 'keywords' param lacks guidance on array vs string, required count, or search semantics.