spherical-memory-mcp

spherical-memory-mcp

Enables AI agents to store and recall memories in a spherical coordinate space with gravitational associations, supporting spatial storage, semantic linking, and human-like forgetting.

Category
Visit Server

README

球状网络标签记忆体系 — MCP Server

一个基于三维球坐标与引力链接的 AI Agent 记忆插件,实现了"球状网络标签记忆体系"论文的核心构想。

任何支持 MCP(Model Context Protocol)的 LLM Agent,接入此插件后即可获得:

  • 空间化记忆存储:每条记忆在闭合球状空间中有唯一的坐标位置
  • 引力联想检索:记忆之间通过语义/情感/因果关联引力自组织为网络
  • 类人遗忘机制:记忆衰减但不删除,可被强关联唤醒
  • 🫀 对话心跳:自动追踪轮次,定时触发记忆固化

快速开始

安装

# 开发安装(推荐)
git clone <this-repo>
cd spherical-memory-mcp
pip install -e .

配置到 Agent

WorkBuddy

编辑 ~/.workbuddy/mcp.json

{
  "mcpServers": {
    "spherical-memory": {
      "command": "/path/to/python",
      "args": ["-m", "spherical_memory.server"],
      "description": "球状网络标签记忆体系"
    }
  }
}

然后在连接器管理中启用(需要点击 信任)。

Claude Desktop

编辑 Claude Desktop 的 MCP 配置:

{
  "mcpServers": {
    "spherical-memory": {
      "command": "python",
      "args": ["-m", "spherical_memory.server"]
    }
  }
}

通用 MCP 客户端

任何支持 MCP stdio 协议的客户端,配置 command + args 同上即可。


工具总览

工具 用途 调用时机
tool_store_memory 存储记忆节点 有意义交互后
tool_recall_by_gravity 引力联想检索 用户提及话题时
tool_recall_by_coordinate 坐标精确查询 知道时间/类型/事件时
tool_register_event 注册事件锚点 新项目/话题出现时
tool_link_memories 手动建立引力链接 发现深层关联时
tool_get_memory_stats 查看记忆概况 了解记忆全景
tool_decay_memories 执行遗忘衰减 每 24 小时
tool_conversation_heartbeat 🫀 对话心跳 每轮对话结束时

🫀 对话心跳机制

每个接入此插件的 Agent 应在每轮对话结束时调用 tool_conversation_heartbeat

心跳工具自动追踪轮次计数,按配置的间隔(默认每 3 轮)返回 consolidate: true。此时 Agent 应回顾本轮对话的关键信息,批量调用 store_memory 写入球状空间。

对话流程:
  第1轮 → heartbeat → consolidate: false
  第2轮 → heartbeat → consolidate: false
  第3轮 → heartbeat → consolidate: true  ← 批量 store_memory
  第4轮 → heartbeat → consolidate: false
  ...

可通过环境变量 SM_HEARTBEAT_INTERVAL 调整间隔(默认 3)。


记忆写入原则

该存什么

  • ✅ 用户分享的重要事实、决定、偏好
  • ✅ Agent 做出的关键决策及其理由
  • ✅ 有情感价值的互动
  • ✅ 新知识的习得
  • ✅ 项目进展的里程碑

不该存什么

  • ❌ 日常问候、简单确认
  • ❌ 重复信息
  • ❌ 临时性技术细节

标签是引力链接的生命线

每条 store_memory 必须提供 3–8 个精准的 semantic_tags。标签质量直接决定引力检索效果。

"用户决定辞去大厂工作开始创业" → ["辞职", "创业", "大厂", "职业转型", "勇气"]

语义相似度引擎

当前默认使用 Jaccard 标签重叠(零外部依赖,速度快),但存在颗粒度瓶颈——语义相近但文字不同的标签(如"论文"vs"学术写作")Jaccard = 0。

预留了可插拔引擎接口services/similarity.py),后续接入 bge-small-zh embedding 后,语义相似度将从"精确匹配"升级为"向量余弦",引力链接质量将质的飞跃。

切换方式:

from spherical_memory.services.similarity import set_similarity_engine, EmbeddingSimilarityEngine
set_similarity_engine(EmbeddingSimilarityEngine(model="BAAI/bge-small-zh-v1.5"))

配置项

环境变量 默认值 说明
SM_DB_PATH ~/.spherical-memory/memory.db 数据库路径
SM_HEARTBEAT_INTERVAL 3 心跳固化间隔(轮)
SM_LINK_THRESHOLD 0.3 引力链接建立阈值
SM_ACTIVATION_THRESHOLD 0.6 引力检索激活阈值
SM_DECAY_RATE 0.95 记忆衰减系数
SM_ENABLE_EMBEDDING false 是否启用 embedding 引擎

架构

src/spherical_memory/
├── server.py              # FastMCP Server 入口,8 个工具注册
├── config.py              # 全局配置
├── db/
│   ├── schema.py          # 5 张表 + 索引
│   └── connection.py      # SQLite 连接管理(WAL 模式)
├── models/
│   ├── memory.py          # MemoryNode
│   ├── event.py           # Event
│   └── gravity_link.py    # GravityLink
└── services/
    ├── memory_service.py  # 记忆 CRUD + 球坐标计算
    ├── event_service.py   # 事件管理
    ├── gravity_service.py # 引力计算 + 链接建立
    ├── recall_service.py  # 引力扩散 + 坐标检索
    ├── decay_service.py   # 质量衰减 + 唤醒
    ├── heartbeat_service.py # 🫀 对话心跳
    └── similarity.py      # 可插拔语义引擎接口

核心理念

Agent的记忆体系,不应当是传统的数据库格式。Agent作为强交互载体,其记忆的录入及读取方式应更类人,这样在双方的交互中才能让自然语言有更高效率。

详见配套论文《球状网络标签记忆体系》。

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured