How to use agentdb for local AI agent memory over MCP

{
  "surface": "mcp.agentdb",
  "package": "npx [email protected]",
  "server": "agentdb/1.3.0",
  "protocol": "2024-11-05",
  "transport": "stdio",
  "tools": ["agentdb_init", "agentdb_insert", "agentdb_insert_batch", "agentdb_search", "agentdb_delete", "reflexion_store", "reflexion_retrieve", "skill_create", "skill_search", "causal_add_edge", "causal_query", "recall_with_certificate", "learner_discover", "db_stats", "learning_start_session"],
  "tool_count": 35,
  "init_times_ms": [10174, 503, 214],
  "call_times_ms": [96, 2, 1],
  "p50_init_ms": 503,
  "p50_call_ms": 2,
  "calls_ok": 5,
  "calls_total": 6,
  "sample_call": {
    "tool": "db_stats",
    "args": {},
    "result_preview": "📊 Database Statistics:

Causal Edges: 0
Experiments: 0
Observations: 0
Episodes: 0
Skills: 0
Reflections: 0
Vectors: 0"
  },
  "notes": "First-run cold start ~10s (better-sqlite3 native module compilation), subsequent runs 200-500ms. agentdb_init has SQL bug: 'no such column: table' — unquoted SQL keyword. db_stats works perfectly. Version skew: npm 3.0.0-alpha.16, serverInfo 1.3.0. Local-only, no external deps."
}