MCP Memory Server
Cross-session persistent memory system using Markdown files and SQLite FTS5 full-text search for AI coding assistants like OpenCode/Claude Code.
README
MCP Memory Server
跨会话持久化记忆系统。Markdown 文件即记忆 + SQLite FTS5 全文搜索。
给 OpenCode / Claude Code 等 AI 编码助手提供跨会话持久记忆能力,会话结束不遗忘。
快速开始
# 一行命令启动
bunx @mimochamber/memory-server
然后在 opencode.json 中注册:
{
"mcp": {
"mcp-memory": {
"type": "local",
"command": ["bunx", "@mimochamber/memory-server"],
"env": {
"OPENCODE_API_URL": "http://127.0.0.1:4096"
}
}
}
}
重启 OpenCode,Agent 就能使用 memory_save / memory_recall 等工具了。
安装
方式一:全局安装(推荐)
bun install -g @mimochamber/memory-server
方式二:项目依赖
# package.json
{
"devDependencies": {
"@mimochamber/memory-server": "github:你的组织/mcp-memory"
}
}
方式三:直接运行(无需安装)
bunx @mimochamber/memory-server
使用
配置到 OpenCode
在 opencode.json(用户级或项目级)中添加 MCP server 配置:
{
"mcp": {
"mcp-memory": {
"type": "local",
"command": ["bunx", "@mimochamber/memory-server"],
"env": {
"OPENCODE_API_URL": "http://127.0.0.1:4096",
"MCP_MEMORY_POLL_INTERVAL": "30000"
}
}
}
}
环境变量
| 变量 | 默认值 | 说明 |
|---|---|---|
MCP_MEMORY_ROOT |
~/.mcp-memory |
记忆存储根目录 |
OPENCODE_API_URL |
http://127.0.0.1:4096 |
OpenCode API 地址(用于轮询 compaction) |
MCP_MEMORY_POLL_INTERVAL |
30000 |
轮询间隔(ms),设为 0 关闭 |
MCP 工具
memory_save
保存一条记忆。写入 .md 文件并同步到 FTS5 搜索索引。
参数:
content: string # 记忆内容(必填)
scope: "global" | "projects" | "sessions" # 作用域
scope_id: string # 项目/会话 ID
type: string # free|memory|notes|compaction
返回: { path, type, indexed_at, deduplicated }
memory_recall
搜索记忆。FTS5 全文搜索 + BM25 排序。
参数:
query: string # 搜索词(必填)
scope?: string # 过滤
limit?: number # 返回条数(默认 10)
返回: [{ path, snippet, score, scope, scope_id, type }]
其他工具
| 工具 | 说明 |
|---|---|
memory_list |
列出记忆条目(支持分页) |
memory_delete |
删除记忆(指定 path) |
memory_stats |
统计信息(总条数、各作用域分布) |
memory_reconcile |
手动触发磁盘文件 ↔ 索引同步 |
工作原理
Agent (LLM) MCP Memory Server 磁盘
│ │ │
│── memory_save ──────▶ │── 写入 .md 文件 ──────────▶ │
│ │── 更新 FTS5 索引 │
│ │ │
│── memory_recall ────▶ │── FTS5 搜索 │
│◀── results (BM25) ──── │ BM25 排序 + 相对分数过滤 │
│ │ │
│ 后台轮询 ──▶│── GET /api/sessions │
│ │── 发现 compaction → 存记忆 │
核心技术
- 存储格式:Markdown 文件(
.md),人类可读可编辑 - 搜索引擎:SQLite FTS5 全文搜索 + BM25 排序
- 查询策略:OR-join(高召回)+ 相对分数过滤(防噪声)
- 增量同步:基于
size-mtimeMsfingerprint 的懒 reconcile - 零源码改动:纯 MCP 协议接入
记忆目录结构
~/.mcp-memory/
├── global/
│ └── MEMORY.md # 跨项目记忆(用户偏好)
├── projects/
│ └── <project-hash>/
│ └── MEMORY.md # 项目级记忆
└── sessions/
└── <session-id>/
├── notes.md # 自由笔记
└── compactions.md # 自动 compaction 摘要(追加模式)
防膨胀机制
① 内容去重
相同作用域下完全重复的内容自动跳过,返回 deduplicated: true。
② Compaction ID 去重
轮询器追踪已处理的 session.id + message.id,避免重复保存。
③ 单文件追加
同一会话的 compaction 摘要追加到 compactions.md,不创建新文件。
开发
# 克隆
git clone https://github.com/你的组织/mcp-memory.git
cd mcp-memory
bun install
# 测试
bun run test.mjs # 冒烟测试
bun run test-dedup.mjs # 去重测试
bun run test-compaction.mjs # Compaction 追加测试
发布到 npm
.\scripts\publish.ps1 # 预览
.\scripts\publish.ps1 -Execute # 正式发布
技术栈
- 运行时: Bun(
bun:sqlite内置) - 协议: MCP(
@modelcontextprotocol/sdk) - 搜索: SQLite FTS5 + BM25
- 存储: Markdown 文件 + SQLite 索引
- 平台: Windows / macOS / Linux
License
MIT
Recommended Servers
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.
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.
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.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
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.
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.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
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.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.