@davidjinguoxu/agentcore

Local-first AI memory engine for multi-agent workflows. 17-tool MCP server. Zero config (LanceDB) or Azure (Cosmos + AI Search).

Any AI tool that speaks MCP — Cursor Agent, Gemini CLI, GitHub Copilot, Claude Code — connects to this server and gains persistent memory, task coordination, and a shared knowledge graph across every session.

Quick start

npx @davidjinguoxu/agentcore

Starts on port 3001. Stores data in ~/.agentcore/ using LanceDB. No account, no API key.

Connect your AI tool

Cursor Agent (~/.cursor/mcp.json):

{
  "mcpServers": {
    "agentcore": {
      "command": "npx",
      "args": ["@davidjinguoxu/agentcore"],
      "env": { "MCP_API_KEY": "your-key" }
    }
  }
}

Gemini CLI (~/.gemini/settings.json):

{
  "mcpServers": {
    "agentcore": {
      "command": "npx",
      "args": ["@davidjinguoxu/agentcore"],
      "env": { "MCP_API_KEY": "your-key" }
    }
  }
}

GitHub Copilot (~/.copilot/mcp-config.json):

{
  "mcpServers": {
    "agentcore": {
      "command": "npx",
      "args": ["@davidjinguoxu/agentcore"],
      "env": { "MCP_API_KEY": "your-key" }
    }
  }
}

The 17 tools

Organized in 5 categories:

| Category | Tools | Layer | |---|---|---| | Memory | log_event, search, deep_search, get_questions | Bronze | | Tasks | start_task, update_task, complete_task, block_task, fail_task, get_tasks | Bronze | | Knowledge | add_entity, get_facts, get_linked | Gold | | Insights | save_lesson, save_decision, ask_team | Bronze→Gold | | Context | warm_start | All layers |

Full reference: docs/TOOLS.md

Three-layer architecture

Bronze  →  raw events (append-only, every AI action)
Silver  →  vector index (semantic search)
Gold    →  knowledge graph (permanent curated facts)

Every piece of data starts in Bronze. Content-rich events get indexed in Silver. Important facts get promoted to Gold via add_entity.

warm_start reads all three layers and assembles a context preamble for the agent.

Full design: docs/ARCHITECTURE.md

Storage backends

| Mode | Condition | Data location | |---|---|---| | LanceDB (default) | No COSMOS_ENDPOINT | ~/.agentcore/ | | Azure Cosmos DB | COSMOS_ENDPOINT + COSMOS_KEY set | Azure cloud | | Azure AI Search | AZURE_SEARCH_ENDPOINT + AZURE_SEARCH_KEY | Azure (Silver layer) |

Environment variables

AGENTCORE_DATA_DIR=~/.agentcore    # local data directory
AGENTCORE_PORT=3001                # HTTP port
MCP_API_KEY=your-key               # auth key for HTTP endpoints

# Cloud (optional)
COSMOS_ENDPOINT=https://...
COSMOS_KEY=...
AZURE_SEARCH_ENDPOINT=https://...
AZURE_SEARCH_KEY=...

Use as a library

npm install @davidjinguoxu/agentcore

import { appendEvent, queryEntities } from '@davidjinguoxu/agentcore';
import type { BronzeRecord, GoldEntity } from '@davidjinguoxu/agentcore/types';

Documentation

| Doc | What it covers | |---|---| | ARCHITECTURE.md | Three-layer data model + all type definitions | | SCHEMA.md | 7 database tables with CREATE TABLE SQL | | MEMORY.md | Hierarchical memory: tiers, time decay, graph traversal | | ACTIVE-SCHEMA.md | Each table's lifecycle + concrete use cases | | TOOLS.md | 17 MCP tools: inputs, outputs, database mapping | | SKILLS.md | 4 skill workflows built on top of tools | | MIGRATION.md | Gap analysis + implementation phases |

npm

@davidjinguoxu/agentcore
https://www.npmjs.com/package/@davidjinguoxu/agentcore