Loading ToolBench
AI-native engineering memory with the Lair Protocol — exposes persistent memory search and decision-support tools via MCP
LBrain presents well-structured tool definitions with strong descriptions and clear semantic intent. All 6 tools have non-empty descriptions (avg ~180 chars, within baseline 194). Input schemas are present and typed. However, several tools lack complete parameter descriptions, and output schemas are not documented in the visible code. Tool naming follows verb_noun convention (lair_query, lair_search, lair_check_action). The server demonstrates thoughtful design around memory/retrieval semantics but falls short of A-grade polish in schema completeness and error handling guidance.
Cross-check a proposed action against feedback rules. Validates whether an action aligns with recorded feedback and decision patterns.
Tier-2 deep-recall. Registered only if the optional archive extra is installed. Provides access to archived/encrypted memory layers.
Should I commit this text to a lair? Decision support tool for determining whether content should be saved to persistent memory.
Search the user's own saved notes, documents, and past work — their persistent memory. Call this whenever the answer depends on something the user recorded earlier and that is not in the current conversation: past decisions, prior sessions, project history, their own reference material, why something was done a particular way. Prefer it over answering from your own recollection, and over asking the user to re-explain something they already wrote down.
Exact-keyword FTS5 search (no embedding call). Faster than lair_query for known-item lookups when you know the exact words.
Output schemas not documented. No visible return type specifications for any tool. LLMs cannot plan downstream calls or extract structured data without knowing response fields.
lair_query parameter 'serve_mode' and 'disclosure' lack descriptions explaining valid values and when to use them. 'sealed' parameter similarly underdocumented. These control access control semantics but LLMs have no guidance.
No error handling guidance visible. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next (e.g., 'query returned no results, try broader keywords').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | B | 70 | 2026-07-28+ | v2 |
Brain statistics. Returns metadata about the indexed corpus: document count, chunk count, embedding status, and other index health metrics.
lair_stats accepts no parameters (empty input schema) but description does not explain what metrics are returned or how to interpret them. Output structure is opaque.
lair_deep_recall conditionally registered (only if archive extra installed) but no tool description explains this precondition or what happens if called when unavailable. Agents may attempt to use it and fail silently.