# Memory MCP — Persistent AI Agent Knowledge Graph

> MCP server that gives AI agents persistent memory using a local knowledge graph. Stores entities, relationships, and observations across sessions for Claude Code.

## Install

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

## Quick Use

Add to your `.mcp.json`:

```json
{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}
```

Restart Claude Code. The agent now has persistent memory across sessions.

## What is Memory MCP?

Memory MCP is an official Model Context Protocol server that provides AI agents with persistent memory through a local knowledge graph. It stores entities, relationships, and observations in a JSON file — enabling agents to remember facts, preferences, and context across conversations.

**Answer-Ready**: Memory MCP is an official MCP server that gives AI agents like Claude Code persistent memory via a local knowledge graph. It stores entities, relationships, and observations in JSON, enabling cross-session context retention.

**Best for**: Claude Code and Cline users who want their AI to remember project context. **Works with**: Any MCP-compatible tool (Claude Code, Cline, Cursor). **Setup time**: Under 1 minute.

## Core Features

### 1. Entity Management
Store and retrieve named entities with types:

```
Agent creates:
  Entity: "Auth Module" (type: component)
  Entity: "Alice" (type: team_member)
  Entity: "PostgreSQL" (type: technology)
```

### 2. Relationships
Connect entities with typed relationships:

```
  "Auth Module" --uses--> "PostgreSQL"
  "Alice" --maintains--> "Auth Module"
  "Auth Module" --depends_on--> "JWT Library"
```

### 3. Observations
Attach facts to entities that accumulate over time:

```
Entity: "Auth Module"
Observations:
  - "Refactored to use JWT in March 2026"
  - "Has 95% test coverage"
  - "Handles 10k req/s in production"
```

### 4. Available Tools

| Tool | Description |
|------|-------------|
| `create_entities` | Add new entities with types |
| `create_relations` | Link entities together |
| `add_observations` | Attach facts to entities |
| `search_nodes` | Full-text search across the graph |
| `open_nodes` | Retrieve specific entities |
| `delete_entities` | Remove entities |
| `delete_observations` | Remove specific facts |
| `delete_relations` | Remove relationships |

### 5. Storage
Data persists in `~/.claude/memory.json` (configurable):

```json
{
  "entities": [
    {
      "name": "Auth Module",
      "entityType": "component",
      "observations": ["Uses JWT tokens", "95% test coverage"]
    }
  ],
  "relations": [
    {
      "from": "Auth Module",
      "to": "PostgreSQL",
      "relationType": "uses"
    }
  ]
}
```

## Use Cases

| Use Case | How |
|----------|-----|
| Project context | Remember architecture decisions across sessions |
| Team info | Store who works on what |
| Preferences | Remember coding style and tool preferences |
| Debug history | Track past issues and their solutions |

## FAQ

**Q: Where is the data stored?**
A: In a local JSON file (default: `~/.claude/memory.json`). Never sent to external servers.

**Q: Does it work with Claude Code?**
A: Yes, it is an official MCP server designed for Claude Code and other MCP-compatible tools.

**Q: How much data can it handle?**
A: Practical limit is thousands of entities. For larger datasets, consider a database-backed MCP server.

## Source & Thanks

> Created by [Anthropic](https://github.com/modelcontextprotocol). Licensed under MIT.
>
> [modelcontextprotocol/servers](https://github.com/modelcontextprotocol/servers) — Part of the official MCP servers collection

<!-- ZH -->

## 快速使用

在 `.mcp.json` 中添加配置，重启 Claude Code 即可使用：

```json
{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}
```

## 什么是 Memory MCP？

Memory MCP 是官方 MCP 服务器，通过本地知识图谱为 AI 代理提供持久记忆。存储实体、关系和观察，跨会话保持上下文。

**一句话总结**：Memory MCP 是官方 MCP 服务器，通过本地知识图谱为 Claude Code 等工具提供跨会话持久记忆。

**适合人群**：希望 AI 记住项目上下文的 Claude Code/Cline 用户。**安装时间**：1 分钟内。

## 核心功能

### 1. 实体管理
创建带类型的命名实体（组件、团队成员、技术等）。

### 2. 关系图谱
用类型化关系连接实体。

### 3. 观察记录
为实体附加事实，随时间积累。

### 4. 全文搜索
搜索整个知识图谱。

## 常见问题

**Q: 数据存在哪？**
A: 本地 JSON 文件，不发送到外部服务器。

**Q: 支持 Claude Code 吗？**
A: 支持，这是官方 MCP 服务器。

## 来源与致谢

> [modelcontextprotocol/servers](https://github.com/modelcontextprotocol/servers) — 官方 MCP 服务器集合, MIT

---
Source: https://tokrepo.com/en/workflows/554c4dc2-663c-489a-b57f-814948e9f714
Author: MCP Hub