Loading ToolBench
Memory with Information-theoretic Relevance Allocation - An MCP memory server that stores, retrieves, and manages memories using HNSW vector search, Cybertron embeddings, and built-in agent-memory continuity
MIRA defines 16 tools with consistent naming (verb_noun pattern) and reasonable descriptions (avg ~80 chars). All tools have input schemas with typed parameters and descriptions. However, output schemas are entirely undocumented, LLMs cannot predict what fields to expect from responses. Parameter descriptions lack actionable constraints (enums, ranges, formats). Error handling guidance is absent. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk stratification (WRITE, DESTRUCTIVE, READ_ONLY). Descriptions are functional but generic, averaging 60 - 100 chars, below the 194-char baseline for A+ tools. Missing dependency hints and recovery guidance.
Archive memories based on age or other criteria
Capture and store agent memory traits and observations
Check for behavioral drift in agent memory
Clear all memories from a specific wing or room
Compress memories to reduce storage while preserving semantic information
Consolidate multiple related memories into a single summary memory
Delete a specific memory by fingerprint ID
Retrieve causal relationships and chains between memories
Output schemas completely undocumented. LLMs cannot predict response structure, field types, or pagination. Forces agents to guess what data is available after each call.
No tool annotations despite explicit risk classification (WRITE, DESTRUCTIVE, READ_ONLY). Destructive tools (clear_memory, delete_memory) lack confirmation/dry-run support. LLMs cannot distinguish safe from irreversible operations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Get current status and statistics of the memory system
Retrieve memories in chronological order with optional filtering
Load a specific memory by its fingerprint ID
Recall agent memory traits and observations with optional filtering
Retrieve memories based on semantic similarity search with optional budget and filters
Perform semantic search across memories with advanced filtering
Store a new memory with optional metadata (wing, room, type, kind)
Update an existing memory's content or metadata
Parameter descriptions lack actionable constraints. 'threshold' (0-1 range) and 'compression_ratio' (0-1 range) have no min/max declarations. 'direction' enum values not listed in description. LLMs cannot validate inputs without explicit constraints.
No error handling guidance. Tools return success/failure but do not explain recovery steps. 'Invalid fingerprint_id' error provides no hint to call search_semantic or recall_memory first.
Descriptions are generic and under 100 chars (baseline 194 chars for A+ tools). Missing WHEN to use each tool and dependencies. E.g., 'recall_memory' vs 'search_semantic' distinction unclear; no guidance on which to call first.