A semantic memory storage and retrieval system with vector embeddings, memory relationships, and usage tracking
Memory Box defines 14 tools with reasonable naming (verb_noun pattern) and mostly complete schemas. However, descriptions are inconsistent in depth, some are generic (e.g., 'Check the health status of the API and database'), and several parameters lack descriptions entirely. Input schemas are present for all tools but vary in completeness. Error handling is minimal; no recovery guidance or actionable error messages visible. Output schemas are partially documented via Pydantic models but not consistently exposed in tool definitions. The server follows basic structure but lacks LLM-optimized descriptions and comprehensive parameter documentation expected of production-grade tools.
Create a new memory bucket for organizing memories
Delete a bucket and optionally its memories
Delete a memory and its relationships
Retrieve a specific memory by ID
Get the processing status of a memory
Get memories related to a specific memory
Get usage metrics for a user
Descriptions lack LLM-optimized guidance. Many are under 100 chars and generic (e.g., 'Check the health status of the API and database'). Should state WHAT, WHEN to use, and any prerequisites. Baseline: 194 chars avg for A+ tools.
No error handling guidance visible. Tools lack recovery hints (e.g., 'If memory not found, try search_memories()'). No categorization of errors as retryable, user-fixable, or fatal.
Destructive operations (delete_memory, delete_bucket) lack confirmation or dry-run support. Agents can accidentally delete data without safeguards.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | <=2025-11-25 | v2 |
Get the current plan for a user
Check the health status of the API and database
List all buckets for a user
List all memories for a user with pagination
Search for memories using semantic similarity and keyword matching
Store a new memory with optional raw content and reference data
Update a user's plan (admin only)
Output schemas not explicitly documented in tool definitions. Pydantic models exist (MemoryResponse, MemoriesListResponse) but are not linked to tool descriptions. LLMs cannot infer response structure.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) visible. MCP spec 2026-07-28 recommends these for agent planning.