memory_strategy
Shipped in v3-alpha (Beta). Interface may still change before 3.0 stable.
Description
Store or recall behavioral strategies — “when X, do Y” patterns learned from experience.
Input schema
{
op: "recall" | "store" | "rate",
recall?: { situation: string },
store?: { trigger; action; context; relatedFacts? },
rate?: { strategyId; outcome: "success" | "failure" },
}Example
// What strategies apply to my current situation?
const strats = await mcp.callTool("memory_strategy", {
op: "recall",
recall: { situation: "Webhook is timing out and customer is escalating" },
});
// Store a new strategy after a successful resolution
await mcp.callTool("memory_strategy", {
op: "store",
store: {
trigger: "Webhook timeout under high traffic",
action: "Increase timeout + enable retry with exponential backoff",
context: "Stripe webhooks specifically — they retry up to 3 days",
relatedFacts: ["fact_8f3e"],
},
});
// Rate a strategy after using it
await mcp.callTool("memory_strategy", {
op: "rate",
rate: { strategyId: "strat_abc", outcome: "success" },
});Backed by
- Procedural memory + prompt evolution — concept
- Memory Worth — strategies are worth-gated like facts