memory_introspect
Shipped in v3-alpha (Beta). Interface may still change before 3.0 stable.
Description
Inspect your own memory health. See what’s strong, what’s weak, what’s missing.
Input schema
{
aspect: "health" | "worth" | "drift" | "types" | "recent_changes" | "coverage" | "gaps" | "strengths",
entity?: string,
limit?: number,
}Example
// Before deciding to recall vs ask the user
const cov = await mcp.callTool("memory_introspect", {
aspect: "coverage",
entity: "deploy-process",
});
// → { coverage: "expert", freshness: "current", factCount: 14, avgWorth: 0.82 }
// Find topics with high recall attempts but low hit rate
const gaps = await mcp.callTool("memory_introspect", {
aspect: "gaps",
limit: 5,
});
// → [{ topic: "kafka-config", recallAttempts: 23, hitRate: 0.13, suggestedAction: "ask_user" }, ...]When to use
Right before a recall call. If coverage is "blind", ask the user
directly. If coverage is "expert" and freshness is "current", trust
the recall result without confirmation.
Backed by
- Meta-cognition — concept + SQL
- Predictive memory — companion zero-LLM feature