Loading ToolBench
Neuroplastic memory engine for Claude Desktop. Ingest conversation text into a memory graph, retrieve relevant memories, and run dream cycles for memory consolidation and pattern detection.
Mind Memory exposes 8 tools with clear verb-noun naming (memory_store, memory_retrieve, memory_dream, etc.). All visible tools have descriptions (10 - 200 chars, within baseline 34 - 392 range). Input schemas are present for 7 of 8 tools with proper type definitions and parameter descriptions. However, tool #8 (memo) has a truncated definition in source, making it unverifiable. Parameter descriptions are generally adequate but lack explicit constraints (enums, ranges, format specs). Output schemas are documented as dicts but lack formal JSON Schema. Error handling is minimal, no recovery guidance or actionable error messages. The server follows basic tool composition patterns but lacks idempotency guarantees and confirmation steps for destructive operations (memory_store, memory_dream, memory_tune).
Tool definition truncated in source code
Compute graph topology diagnostics: orphans, density, clustering, hit rates.
Run a dream cycle: NREM consolidation, REM exploration, threat simulation, pattern detection.
Return the last dream cycle's narrative in human-readable text.
Retrieve relevant memories for a query. Returns formatted context for prompt injection.
Return current graph statistics.
Ingest conversation text into the memory graph. Extracts concepts and relationships, then persists.
Tool #8 (memo) definition truncated in source code, cannot verify schema, description, or parameters. Unverifiable tool capped at 0.
No input validation or constraint documentation. Parameters like 'k' (number of results) and 'alpha' (blend weight) lack min/max bounds. LLMs may pass invalid values (k=0, alpha=2.5) without guidance.
Destructive operations (memory_store, memory_dream, memory_tune) lack confirmation or dry-run support. No idempotency guarantees, repeated calls may produce duplicate side effects.
Error handling is minimal. Tools return dicts with success counts but no guidance on partial failures, retryability, or recovery steps. E.g., memory_store returns node counts but not which concepts failed extraction.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
Tune a config parameter at runtime. Use 'section.key' format (e.g. 'dream.rem_jump_probability').
Output schemas documented as Python dicts but not formally specified in JSON Schema. Downstream tools cannot validate response structure. memory_dream returns nested dicts (dream_paths, threat_flags, suggestions) without field type guarantees.