# DuckDB Graph Memory MCP — File-Backed Memory DB

> DuckDB Graph Memory MCP is a fork of the MCP memory server that stores the knowledge graph in a DuckDB file for SQL queries and scaling beyond JSON.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

1. Add to `claude_desktop_config.json`:
   ```json
   {
     "mcpServers": {
       "graph-memory": {
         "command": "npx",
         "args": ["-y", "@izumisy/mcp-duckdb-memory-server"],
         "env": {
           "MEMORY_FILE_PATH": "/path/to/your/memory.data"
         }
       }
     }
   }
   ```
2. Restart Claude Desktop / your MCP client.
3. Verify: store 3 facts, then query them back and confirm the file path is a DuckDB database file.

## Intro

DuckDB Graph Memory MCP is a fork of the MCP memory server that stores the knowledge graph in a DuckDB file for SQL queries and scaling beyond JSON.

- **Best for:** agent builders who need a self-hosted memory graph that can grow beyond JSON and remain queryable
- **Works with:** MCP clients (Claude Desktop), Node.js npx runner, a writable filesystem path for the DuckDB file
- **Setup time:** 10-20 minutes

## Practical Notes

- Quant: memory is stored in a DuckDB file, so you can backup/rotate it like any other database artifact.
- Quant: treat memory writes as an explicit policy—capture only verified facts to keep retrieval precision high.

## Why it matters

JSON-based memory servers often hit scaling and query flexibility limits. DuckDB-backed storage gives you SQL ergonomics and stronger integrity for long-lived agent memory.

- Simple npx-based setup keeps integration lightweight.
- File-backed storage enables backups, migrations, and audits with standard tooling.
- The README explains why DuckDB helps (performance, scalability, query flexibility, transactions).

## Rollout pattern

- Start with a dedicated memory file per project to avoid cross-project contamination.
- Define a memory write policy: only write decisions, constraints, and stable preferences.
- Schedule periodic cleanup or consolidation so the knowledge graph stays useful over time.

## Watchouts

A local memory file can still leak secrets if your agent writes them. Treat the file as sensitive and keep it out of synced folders or shared drives.

### FAQ

**Q: Is this the official MCP memory server?**
A: It is a fork of the official knowledge graph memory server with DuckDB storage changes.

**Q: Where is data stored?**
A: In a DuckDB database file at `MEMORY_FILE_PATH`.

**Q: How do I keep it clean?**
A: Write only verified facts and run periodic consolidation/cleanup so retrieval stays high-signal.

## Source & Thanks

> Source: https://github.com/IzumiSy/mcp-duckdb-memory-server
> License: MIT
> GitHub stars: 55 · forks: 12

---

<!-- ZH -->

## 快速使用

1. 写入 `claude_desktop_config.json`：
   ```json
   {
     "mcpServers": {
       "graph-memory": {
         "command": "npx",
         "args": ["-y", "@izumisy/mcp-duckdb-memory-server"],
         "env": {
           "MEMORY_FILE_PATH": "/path/to/your/memory.data"
         }
       }
     }
   }
   ```
2. 重启 Claude Desktop / MCP 客户端。
3. 验证：先写入 3 条事实，再检索回来，并确认该路径对应的是 DuckDB 数据库文件。

## 简介

DuckDB Graph Memory MCP 是 MCP 记忆服务器的一个分支版本，把知识图谱记忆存到 DuckDB 文件里，支持更强的 SQL 查询与更好的扩展性，适合长期记忆变大后的检索与治理。

- **适合谁：** 需要自托管知识图谱记忆，并希望记忆规模变大后仍可查询/治理的 Agent 构建者
- **可搭配：** MCP 客户端（Claude Desktop 等）、Node.js npx 运行器、可写的本地文件路径（DuckDB 文件）
- **准备时间：** 10-20 分钟

## 实战建议

- 量化信息：记忆以 DuckDB 文件形式存储，可像数据库一样做备份/轮换与版本管理。
- 量化信息：把写入当作策略：只写已验证事实，才能保持检索精度。

## 为什么值得收录

JSON 记忆服务在规模变大时容易遇到性能和查询灵活性问题。DuckDB 作为底座能带来 SQL 的可操作性与更强的数据完整性，更适合长期记忆。

- npx 方式接入简单轻量，集成成本低。
- 文件型数据库便于备份、审计与迁移。
- README 解释了 DuckDB 的优势（性能/扩展性/事务/查询灵活性）。

## 落地路径

- 每个项目独立一份 memory 文件，避免跨项目串线。
- 定义写入策略：只记录决策、约束与稳定偏好。
- 定期做清理/归并，保持知识图谱长期高信噪比。

## 注意事项

本地记忆文件依然可能泄露秘密（如果 agent 写入了）。把它当敏感数据，避免放到同步盘或共享盘。

### FAQ

**这是官方版本吗？**
答：它是官方知识图谱记忆服务的 fork，主要改为 DuckDB 存储。

**数据存在哪里？**
答：存到 `MEMORY_FILE_PATH` 指向的 DuckDB 数据库文件里。

**如何保持干净？**
答：只写已验证事实，并定期做归并/清理，避免检索变噪音。

## 来源与感谢

> Source: https://github.com/IzumiSy/mcp-duckdb-memory-server
> License: MIT
> GitHub stars: 55 · forks: 12


---
Source: https://tokrepo.com/en/workflows/duckdb-graph-memory-mcp-file-backed-memory-db
Author: MCP Hub