⚡ agentvault-cache

Smart caching for MCP tool results — reduce latency, cut costs, eliminate redundant calls.

Why?

MCP tool calls can be slow and expensive. If your agent calls the same tool with the same arguments multiple times, you're wasting time and money. agentvault-cache wraps any tool call with intelligent TTL-based caching.

Quick Start

import { createCache } from 'agentvault-cache';

const cache = createCache({ ttlMs: 60000 }); // 1 minute TTL

// Wrap any async function:
const result = await cache.wrap('search:test', () => callTool('search', { query: 'test' }));

// Second call with same key → instant, from cache
const cached = await cache.wrap('search:test', () => callTool('search', { query: 'test' }));

Features

| Feature | Description | |---------|-------------| | TTL-based expiry | Entries auto-expire after configurable duration | | LRU eviction | Least recently used entries evicted when cache is full | | Per-entry TTL | Override TTL for specific entries | | Hit rate stats | Monitor cache effectiveness with .stats() | | Zero dependencies | Pure JavaScript, no external deps |

API

createCache(options)

| Option | Type | Default | Description | |--------|------|---------|-------------| | ttlMs | number | 300000 | Default TTL in milliseconds | | maxSize | number | 1000 | Max entries before LRU eviction |

cache.wrap(key, fn)

Cache the result of fn() under key. If a non-expired entry exists, returns it immediately.

cache.get(key)

Get a cached entry. Returns undefined if not found or expired.

cache.set(key, value, ttlMs?)

Manually set a cache entry with optional per-entry TTL.

cache.invalidate(key)

Remove a specific entry.

cache.clear()

Remove all entries.

cache.stats()

Returns { hits, misses, hitRate, size }.

Part of AgentVault

🏴‍☠️ AgentVault — The essential toolkit for AI agents.

| Package | What it does | |---------|-------------| | agentvault-tools | 🛠️ 9-tool MCP starter kit (search, memory, calc) | | agentvault-memory | 🧠 Local-first persistent memory | | agentvault-guard | 🛡️ Tool poisoning detection | | agentvault-gateway | 🔌 MCP proxy with audit logging | | agentvault-score | 📊 Quality scoring (A+ to F) | | agentvault-retry | 🔄 Resilient calls with backoff | | agentvault-cache | ⚡ Smart result caching (you are here) |

MIT License