Loading ToolBench
Drop-in memory for Claude Code, OpenClaw, and any MCP-compatible agent. Provides hybrid search (semantic + full-text), governance, and recall tools for AI agent memory management.
mind-mem provides 11 tools with clear action verbs and mostly complete schemas. Tool names follow verb_noun conventions (recall, capture, propose_update, etc.). Most tools have descriptions (194 - 370 chars) and documented input parameters with types. However, output schemas are not visible in the source code provided, and some descriptions lack depth around when to use a tool versus alternatives. Error handling guidance is minimal. Tools are well-composed (each does one thing), but parameter descriptions could be more prescriptive about formats, ranges, and constraints. STDIO-only transport is a hard blocker for remote accessibility.
Approve and apply a proposed governance update. Finalizes a pending proposal and commits it to the memory corpus.
Create a backup of the memory corpus. Exports the entire knowledge base to a portable format.
Capture and store new memories, decisions, or context into the knowledge base. Creates timestamped entries with tags and metadata.
Perform hybrid search combining semantic similarity and full-text search with Reciprocal Rank Fusion (RRF) ranking.
Propose a governance update to the memory corpus. Initiates a multi-step approval workflow for memory modifications.
Query the memory corpus using hybrid search (semantic + full-text retrieval, RRF fusion). Returns ranked results with relevance scores and context windows.
Restore memory corpus from a backup. Imports previously exported knowledge base.
Output schemas are not documented in the provided source. LLMs cannot reason about what fields to expect from recall, capture, or scan responses, forcing them to guess or inspect responses blindly.
Parameter descriptions lack prescriptive guidance on formats, ranges, and constraints. E.g., 'search query' does not explain if natural language is preferred, max length, or special syntax. 'scope' in scan has enum options but no guidance on when to use 'all' vs 'active' vs 'recent'.
No error handling guidance visible. Tools like approve_apply and rollback_proposal mutate state but provide no recovery hints if approval fails, proposal is not found, or rollback conflicts arise.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | C | 65 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 16 | - | v1 |
Rollback a governance proposal. Reverts a pending or applied update and restores the previous state.
Scan the memory corpus for integrity issues, contradictions, drift, and dead decisions. Reports structural and logical inconsistencies.
Build or rebuild FTS5 full-text search index for efficient retrieval. Manages the search index for the memory corpus.
Validate Python code stored in memories against structural checks (imports, syntax, type hints). Runs 74+ structural checks on current workspace.
hybrid_search appears to duplicate functionality of recall (both perform search, recall just includes additional parameters like 'backend' and 'format'). No clear guidance on when to use hybrid_search vs recall.
Tool descriptions do not explain output structure or pagination behavior. Tools returning lists (recall, scan) do not document whether results are paginated, how many results are returned by default, or how to fetch more.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) visible in definitions. This prevents MCP clients from displaying appropriate warnings or confirmations for irreversible operations like rollback_proposal or restore.