MemoryWeaver

统一 Agent 记忆中枢 / Universal Agent Memory Hub

English | 中文

---

English

Overview

MemoryWeaver is a unified memory management MCP Server for AI coding agents. It solves the "memory island" problem where switching between tools (Claude Code, Cursor, Trae, Hermes, etc.) loses context and requires re-configuring your preferences.

Features

• Multiple Memory Types: user / project / feedback / reference / temporary
• Semantic Search: Vector search powered by ChromaDB
• Full-text Search: SQLite FTS5 for exact matching
• Auto-extraction: Extract memories from conversations using DeepSeek API
• File Injection: Generate context files for Cursor / Claude Code / generic agents

Installation

# Clone the repository
git clone https://github.com/JianFeiGan/memoryweaver.git
cd memoryweaver

# Install dependencies
uv sync

Configuration

Copy config.example.yaml to config.yaml and fill in your API keys:

cp config.example.yaml config.yaml

Environment variable substitution is supported using ${ENV_VAR} syntax:

api:
  embedding:
    api_key: "${OPENAI_API_KEY}"
  deepseek:
    api_key: "${DEEPSEEK_API_KEY}"

Usage

Start the MCP Server

uv run python -m src.server

Configure in Claude Code

Add to your MCP settings:

{
  "mcpServers": {
    "memoryweaver": {
      "command": "uv",
      "args": ["run", "python", "-m", "src.server"],
      "cwd": "/path/to/MemoryWeaver"
    }
  }
}

Configure in Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "memoryweaver": {
      "command": "uv",
      "args": ["run", "python", "-m", "src.server"],
      "cwd": "/path/to/MemoryWeaver"
    }
  }
}

MCP Tools

Tool	Description
save_memory	Save a memory with type, content, and metadata
search_memory	Search memories (semantic + full-text)
list_memories	List memories with optional filters
get_project_context	Get formatted project context for injection
extract_from_conversation	Auto-extract memories from conversation

Testing

uv run pytest tests/ -v

Architecture

MemoryWeaver/
├── src/
│   ├── server.py              # MCP Server entry point
│   ├── memory_engine.py       # Core memory engine
│   ├── auto_extractor.py      # Auto-extraction module
│   ├── search_engine.py       # Search engine (semantic + full-text)
│   ├── file_injector.py       # File injection module
│   ├── models.py              # Data models
│   └── config.py              # Configuration
├── tests/                     # Test suite
├── pyproject.toml             # Project config
└── config.yaml                # Runtime config

Tech Stack

• Python 3.11+
• MCP SDK - Model Context Protocol
• SQLite - Metadata storage
• ChromaDB - Vector storage
• OpenAI API - Embeddings
• DeepSeek API - Auto-extraction

License

MIT

---

中文

概述

MemoryWeaver 是一个统一的 AI 编码 Agent 记忆管理 MCP Server。它解决了"记忆孤岛"问题——在 Claude Code、Cursor、Trae、Hermes 等工具之间切换时丢失上下文，需要重新配置偏好设置。

功能特性

• 多种记忆类型: 个人记忆 / 项目记忆 / 反馈记忆 / 引用记忆 / 临时记忆
• 语义搜索: 基于 ChromaDB 的向量搜索，理解查询意图
• 全文搜索: 基于 SQLite FTS5 的精确匹配
• 自动提炼: 使用 DeepSeek API 从对话中自动提取记忆
• 文件注入: 为 Cursor / Claude Code / 通用 Agent 生成上下文文件

安装

# 克隆仓库
git clone https://github.com/JianFeiGan/memoryweaver.git
cd memoryweaver

# 安装依赖
uv sync

配置

复制 config.example.yaml 为 config.yaml，填入你的 API Key:

cp config.example.yaml config.yaml

支持环境变量替换，使用 ${ENV_VAR} 语法:

api:
  embedding:
    api_key: "${OPENAI_API_KEY}"
  deepseek:
    api_key: "${DEEPSEEK_API_KEY}"

使用方法

启动 MCP Server

uv run python -m src.server

在 Claude Code 中配置

添加到 MCP 设置:

{
  "mcpServers": {
    "memoryweaver": {
      "command": "uv",
      "args": ["run", "python", "-m", "src.server"],
      "cwd": "/path/to/MemoryWeaver"
    }
  }
}

在 Cursor 中配置

添加到 .cursor/mcp.json:

{
  "mcpServers": {
    "memoryweaver": {
      "command": "uv",
      "args": ["run", "python", "-m", "src.server"],
      "cwd": "/path/to/MemoryWeaver"
    }
  }
}

MCP 工具

工具	说明
save_memory	保存记忆，支持类型、内容和元数据
search_memory	搜索记忆 (语义搜索 + 全文搜索)
list_memories	列出记忆，支持按类型/项目过滤
get_project_context	获取格式化的项目上下文，用于文件注入
extract_from_conversation	从对话中自动提取记忆

测试

uv run pytest tests/ -v

项目架构

MemoryWeaver/
├── src/
│   ├── server.py              # MCP Server 入口
│   ├── memory_engine.py       # 核心记忆引擎
│   ├── auto_extractor.py      # 自动提炼模块
│   ├── search_engine.py       # 搜索引擎 (语义 + 全文)
│   ├── file_injector.py       # 文件注入模块
│   ├── models.py              # 数据模型
│   └── config.py              # 配置管理
├── tests/                     # 测试套件
├── pyproject.toml             # 项目配置
└── config.yaml                # 运行时配置

技术栈

• Python 3.11+
• MCP SDK - Model Context Protocol
• SQLite - 元数据存储
• ChromaDB - 向量存储
• OpenAI API - 文本嵌入
• DeepSeek API - 自动提炼

许可证

MIT