给 AI 助手装上“第二大脑”:基于 Mem0 与 MCP 的本地轻量级记忆方案

# 创建并进入专属工作目录
mkdir -p ~/Documents/Memory0/qdrant
cd ~/Documents/Memory0

# 创建虚拟环境（建议 Python >= 3.10）
uv venv .venv
source .venv/bin/activate

uv pip install mem0ai qdrant-client mcp

#!/usr/bin/env python3
import os
from dataclasses import dataclass
from pathlib import Path
from mcp.server.mcpserver import MCPServer
from mem0 import Memory

# 确保本地回环地址直连，不受全局代理干扰
os.environ["NO_PROXY"] = "127.0.0.1,localhost"
os.environ["no_proxy"] = "127.0.0.1,localhost"


@dataclass
classConfig:
    storage_dir: str = str(Path.home() / "Documents/Memory0/qdrant")

# 向量嵌入配置
    embed_base_url: str = os.getenv("EMBED_BASE_URL", "https://api.example.com/v1")
    embed_api_key: str = os.getenv("EMBED_API_KEY", "your-embed-key")
    embed_model: str = "Qwen3-Embedding-8B"
    embed_dims: int = 1024

# 实体与事实抽取 LLM 配置
    llm_base_url: str = os.getenv("LLM_BASE_URL", "http://127.0.0.1:8317/v1")
    llm_api_key: str = os.getenv("LLM_API_KEY", "your-llm-key")
    llm_model: str = "your-llm-model"


cfg = Config()

memory = Memory.from_config({
"vector_store": {
"provider": "qdrant",
"config": {
"collection_name": "developer_memories",
"path": cfg.storage_dir,
"embedding_model_dims": cfg.embed_dims,
        },
    },
"llm": {
"provider": "openai",
"config": {
"model": cfg.llm_model,
"api_key": cfg.llm_api_key,
"openai_base_url": cfg.llm_base_url,
        },
    },
"embedder": {
"provider": "openai",
"config": {
"model": cfg.embed_model,
"api_key": cfg.embed_api_key,
"openai_base_url": cfg.embed_base_url,
"embedding_dims": cfg.embed_dims,
        },
    },
})

mcp = MCPServer("local-mem0")


@mcp.tool()
defsearch_memories(query: str, top_k: int = 5) -> str:
"""按问题检索历史记忆、偏好及环境配置"""
    results = memory.search(query, limit=top_k)
    items = [r["memory"] for r in results.get("results", [])]
return"\n".join(items) if items else"未找到相关记忆。"


@mcp.tool()
defremember_fact(text: str) -> str:
"""将一条明确的事实、配置或偏好存入本地记忆库"""
    memory.add(text)
return"已保存至本地向量库。"


if __name__ == "__main__":
    mcp.run()

{
"mcpServers":{
    "local-mem0":{
        "command":"/Users/xxxx/Documents/Memory0/.venv/bin/python",
        "args":["/Users/xxxx/Documents/Memory0/server.py"],
        "env":{
            "NO_PROXY":"127.0.0.1,localhost",
            "PYTHONUNBUFFERED":"1"
        }
    }
}
}

tar -czf ~/Documents/Memory0_backup_$(date +%Y%m%d).tar.gz -C ~/Documents/Memory0 qdrant