Loading ToolBench
Brain-inspired memory system for AI Agents — MCP Server for Claude Code & Codex CLI
Cell-mem has 17 tools with reasonable naming (verb_noun pattern) and descriptions, but critical gaps undermine quality. Most tools lack visible input schemas in the provided code, only parameter names and types are inferred from the tool list. No output schemas are documented. Many descriptions are generic or lack actionable context (e.g., 'Retrieve memories' without explaining when to use it vs. memory_recall). Several tools (memory_associate, memory_forget, memory_consolidate, memory_verify, memory_reflect, memory_replay, memory_hypothesis_feedback, memory_creative_pool, memory_check_environment) are registered in stubs.py with minimal implementation, making it impossible to verify actual schema compliance. Error handling is not visible in the provided code. Security considerations (secret injection, permission gates) are absent from tool definitions.
Create associations between memories in the knowledge graph
Check current environment state and capture snapshots for condition verification
Check for conflicts in stored user preferences
Trigger consolidation of episodic memories into semantic knowledge
Access the creative pool of generated hypotheses and scenarios
Extract user preferences from memories using the preference pipeline
Mark a memory for forgetting or decay in the consolidation process
Retrieve stored user preferences from the preference store
Input schemas not visible in source code. Tool definitions inferred from parameter lists only; actual JSON Schema registration cannot be verified.
Output schemas completely undocumented. No tool describes what fields it returns, data types, or structure. LLMs cannot plan downstream calls or extract required data.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | <=2025-11-25 | v2 |
Provide feedback on hypotheses generated during replay
Inject a user preference into the preference store
Retrieve memories from the Cell-mem system using semantic search across episodic and semantic layers
Record feedback on user preferences for refinement
Trigger reflection engine to generate insights and patterns from memories
Trigger generative replay to prevent catastrophic forgetting
Write a memory to the Cell-mem brain-inspired memory system. Memories are routed to working, episodic, or semantic layers based on memory_type. Semantic memories should have higher confidence (0.5+) for facts and knowledge; episodic for interaction records; working for transient task context.
Get status report of the Cell-mem memory system including memory layer statistics and system health
Verify semantic memories against current environment conditions using falsifiable conditions
Stub implementations in stubs.py (8 tools) lack real logic and error handling. Cannot verify actual behavior, parameter validation, or recovery guidance.
Generic descriptions lack actionable context. 'Retrieve memories' (memory_recall) does not explain when to use it vs. memory_status or memory_extract_preferences. No dependency hints or prerequisites stated.
No error handling or recovery guidance visible. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next.