open-memory

MCP文档知识

提供方 DanWahlin

文档知识 stdio · 本地进程 API Key 社区 / 第三方 ClaudeCursorVS CodeCodex
查看接入方式

open-memory is an MCP server that provides AI tools with access to a personal knowledge base, enabling knowledge sharing across multiple tools through a single protocol, supporting both local and remote access.

一句话选型结论

按资料字段整理

open-memory 属于文档知识类 MCP 服务器,stdio · 本地进程,API Key,可直接接入 Claude、Cursor、VS Code 等,有可直接执行的安装命令。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类文档知识 传输方式stdio · 本地进程 鉴权API Key 部署方式本地部署
GitHub 星标 1 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

文档知识
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

Local Deployment Knowledge Base Integration AI Tool Extension TypeScript MCP Protocol Local Storage Semantic Search

支持客户端

适用场景

Local DeploymentKnowledge Base IntegrationAI Tool Extension

安装与配置

stdio · 本地进程 API Key
客户端在本机拉起子进程,数据不出本机
node /path/to/open-memory/dist/stdio.js

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    stdio 服务在本地起进程,需要先装好 node,再按安装命令拉取依赖。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
文档检索问答

在你的文档库里按问题找答案。

知识库读写

把新资料写进知识库,随用随取。

内容整理

把零散资料归纳成结构化笔记。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude、Cursor、VS Code,换客户端不用重新研究配置。

先想清楚的地方

  • stdio 方式要求本地装好运行时(如 Node.js),换机器要重新装一遍。
  • 使用前要准备好API Key对应的凭据,并按最小权限配置。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在文档知识场景里让模型直接调用外部能力的人。
  • 已经在用 Claude、Cursor 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 完全不想在本地装环境的人。
  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
open-memory当前 未标注 stdio · 本地进程 本地部署 有 2026-05-18
dify 未标注 stdio · 本地进程 本地部署 有 2026-04-11
open-webui 未标注 stdio · 本地进程 本地部署 有 2026-04-11
JeecgBoot 未标注 stdio · 本地进程 本地部署 有 2026-04-11
Cherry Studio 未标注 未标注 未标注 无 2026-05-17
mempalace 未标注 stdio · 本地进程 本地部署 有 2026-04-11

服务器档案

文档知识
字段来自项目仓库与公开索引,中文说明由本站标注
提供方DanWahlin上游标注的维护方 DanWahlin
分类文档知识本站按上游分类展示
连接方式stdio · 本地进程客户端在本机拉起子进程,数据不出本机
鉴权方式API Key远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · node本地安装时用到的包标识与运行时
部署类型本地部署上游给出的部署形态
最近核验2026-05-18上游最后核验时间

详细介绍

8.8 KB
来自项目仓库与上游目录的原文,未做删改

open-memory

One brain, every AI. An MCP server that gives any AI tool (Copilot, Claude, Codex, Cursor, ChatGPT) access to your personal knowledge base through a single open protocol.

Your memory lives as plain markdown files. QMD indexes and searches them (BM25 + vector embeddings). open-memory wraps it all in an MCP server that any client can connect to. No database. No SaaS. Files are the source of truth.```mermaid
graph LR
Copilot -->|MCP| OM[open-memory<br/>stdio or SSE]
Claude_Code[Claude Code] -->|MCP| OM
Cursor -->|MCP| OM
Codex -->|MCP| OM
ChatGPT -->|MCP| OM
OM -->|QMD| FS[memory/.md<br/>MEMORY.md<br/>notes/.md]

style OM fill:#2d333b,stroke:#539bf5,color:#adbac7
style FS fill:#2d333b,stroke:#57ab5a,color:#adbac7

&gt; Transport: local (stdio) or remote (SSE over Tailscale/tunnel)

## Why

Every AI tool has its own memory silo. Copilot doesn&#x27;t know what you told Claude. Cursor doesn&#x27;t remember what Codex learned. You re-explain context every time you switch tools.

open-memory fixes this: one directory of markdown files, one search index, one MCP server. Any tool that speaks MCP gets your full context.

Cost: **$0.** Self-hosted, no external services.

## Quick Start

### Prerequisites

- **Node.js** 20+
- **QMD** (the search/indexing engine)

### Install QMD```bash
# Using bun (fastest)
bun install -g qmd

# Or npm
npm install -g qmd

Set Up Your Memory

If you're starting fresh:```bash
mkdir -p ~/memory/memory
echo "# My Knowledge Base" > ~/memory/MEMORY.md
echo "# $(date +%Y-%m-%d)" > ~/memory/memory/$(date +%Y-%m-%d).md


If you already use OpenClaw, your memory is at `~/.openclaw/` (MEMORY.md + memory/*.md). Point open-memory there.

### Index Your Files with QMD```bash
# Create a collection pointing to your memory directory
qmd collection add ~/memory --name my-memory --mask &quot;**/*.md&quot;

# Generate vector embeddings (needed for semantic search)
# Requires OPENAI_API_KEY or a local embedding model
qmd embed

QMD supports multiple embedding providers. See QMD docs for configuration.

Install & Run open-memory```bash

Clone

git clone https://github.com/DanWahlin/open-memory.git
cd open-memory

Install & build

npm install
npm run build

Configure (optional)

cp .env.example .env

Edit .env to set QMD_BIN path, MEMORY_DIR, auth token, etc.

Run (HTTP/SSE mode for remote clients)

npm start

Or run in stdio mode (for local clients)

npm run start:stdio


## Connecting AI Tools

### GitHub Copilot CLI

The fastest way to connect. Run `copilot` in your terminal, then:```
/mcp add

Fill in the interactive form:

Field Value
Server Name open-memory
Server Type STDIO (press the corresponding number)
Command node /path/to/open-memory/dist/stdio.js
Environment Variables {&quot;QMD_BIN&quot;: &quot;/path/to/qmd&quot;, &quot;MEMORY_DIR&quot;: &quot;/path/to/your/memory&quot;}
Tools *

Press Ctrl+S to save. The server is available immediately, no restart needed.

You can also edit the config file directly at ~/.copilot/mcp-config.json:```json
{
"mcpServers": {
"open-memory": {
"type": "local",
"command": "node",
"args": ["/path/to/open-memory/dist/stdio.js"],
"env": {
"QMD_BIN": "/path/to/qmd",
"MEMORY_DIR": "/path/to/your/memory"
},
"tools": ["*"]
}
}
}


Manage your server with `/mcp show`, `/mcp edit open-memory`, or `/mcp delete open-memory`.

For remote access (SSE), use type `&quot;http&quot;` or `&quot;sse&quot;` with your server URL instead.

### Claude Code```bash
claude mcp add open-memory -- node /path/to/open-memory/dist/stdio.js

Or add to your MCP config:```json
{
"mcpServers": {
"open-memory": {
"command": "node",
"args": ["/path/to/open-memory/dist/stdio.js"],
"env": {
"QMD_BIN": "/path/to/qmd",
"MEMORY_DIR": "/path/to/your/memory"
}
}
}
}


### Cursor

Add to your Cursor MCP settings (Settings &gt; MCP Servers):```json
{
  &quot;mcpServers&quot;: {
    &quot;open-memory&quot;: {
      &quot;command&quot;: &quot;node&quot;,
      &quot;args&quot;: [&quot;/path/to/open-memory/dist/stdio.js&quot;],
      &quot;env&quot;: {
        &quot;QMD_BIN&quot;: &quot;/path/to/qmd&quot;,
        &quot;MEMORY_DIR&quot;: &quot;/path/to/your/memory&quot;
      }
    }
  }
}

Codex CLI

Add to ~/.codex/config.toml:```toml
[mcp_servers.open-memory]
command = 'node'
args = ['/path/to/open-memory/dist/stdio.js']

[mcp_servers.open-memory.env]
QMD_BIN = '/path/to/qmd'
MEMORY_DIR = '/path/to/your/memory'


### VS Code (Copilot Chat)

Add to `.vscode/mcp.json` in your workspace (or `~/.vscode/mcp.json` globally):```json
{
  &quot;servers&quot;: {
    &quot;open-memory&quot;: {
      &quot;command&quot;: &quot;node&quot;,
      &quot;args&quot;: [&quot;/path/to/open-memory/dist/stdio.js&quot;],
      &quot;env&quot;: {
        &quot;QMD_BIN&quot;: &quot;/path/to/qmd&quot;,
        &quot;MEMORY_DIR&quot;: &quot;/path/to/your/memory&quot;
      }
    }
  }
}

Remote Access (SSE mode)

Run the HTTP server on a machine with your memory files:```bash
OPEN_MEMORY_TOKEN=your-secret-token npm start


Connect from any MCP client that supports SSE or HTTP transport:```
URL: http://&lt;your-server&gt;:3838/sse
Auth: Bearer your-secret-token

For GitHub Copilot CLI remote access, use /mcp add with type HTTP or SSE and paste the URL.

Over Tailscale, use your Tailscale IP. Over the internet, put it behind a reverse proxy with HTTPS.

Tools

open-memory exposes 7 MCP tools:

Tool Description
search_memory Semantic, keyword, or hybrid search across your knowledge base
read_memory Read a specific memory file by path
write_memory Append to today's daily note (or a specific file)
list_memories List memory files in a directory
browse_recent Preview recent daily notes
get_document Get any indexed document by path or QMD docid
memory_status Show index health, collections, and document counts

Configuration

All configuration via environment variables (or .env file):

Variable Default Description
QMD_BIN qmd Path to QMD binary
MEMORY_DIR ~/.openclaw Root directory for memory files
DAILY_NOTES_DIR memory Subdirectory for daily notes (within MEMORY_DIR)
MAIN_MEMORY_FILE MEMORY.md Main memory file name
QMD_COLLECTION (empty) QMD collection to search (empty = all)
PORT 3838 HTTP server port (SSE mode)
OPEN_MEMORY_TOKEN (empty) Bearer token for auth (empty = no auth)

QMD Setup Guide

QMD is the search engine behind open-memory. It indexes markdown files and provides BM25 (keyword), vector (semantic), and hybrid search.

Collections

QMD organizes files into collections. Create one for your memory:```bash

Index a directory of markdown files

qmd collection add ~/my-notes --name notes --mask "*/.md"

List your collections

qmd collection list

Re-index after adding files

qmd update


### Embeddings

For semantic search, QMD needs vector embeddings:```bash
# Set your API key (OpenAI, or compatible provider)
export OPENAI_API_KEY=sk-...

# Generate embeddings
qmd embed

# Check status
qmd status

Search Modes```bash

Keyword search (fast, exact matching)

qmd search "project decisions"

Semantic search (finds conceptually related content)

qmd vsearch "what did we decide about the architecture"

Hybrid (best quality, combines both + reranking via local LLM)

qmd query "why did we choose Postgres over MongoDB"


## Architecture```mermaid
graph TB
    Client[MCP Client&lt;br/&gt;Copilot · Claude · Cursor · Codex]
    Client --&gt;|MCP Protocol&lt;br/&gt;stdio or SSE| Server

    subgraph Server[open-memory server]
        search_memory --&gt;|vsearch / search / query| QMD
        read_memory --&gt; FS_read[fs.readFileSync]
        write_memory --&gt; FS_append[fs.appendFileSync]
        get_document --&gt;|get| QMD
        list_memories --&gt; FS_readdir[fs.readdirSync]
        browse_recent --&gt; FS_readdir
        memory_status --&gt;|status| QMD
    end

    Server --&gt; FileSystem

    subgraph FileSystem[File System]
        MEMORY[MEMORY.md&lt;br/&gt;curated long-term knowledge]
        Daily[memory/YYYY-MM-DD.md&lt;br/&gt;daily notes]
        Other[*.md&lt;br/&gt;any indexed markdown]
    end

    style Client fill:#2d333b,stroke:#539bf5,color:#adbac7
    style Server fill:#1c2128,stroke:#539bf5,color:#adbac7
    style FileSystem fill:#1c2128,stroke:#57ab5a,color:#adbac7

Key design decisions:
- Files are the source of truth, not a database
- QMD handles all indexing and search (BM25 + vector)
- Zero external services required
- Two transport modes: stdio (local) and SSE (remote)
- Auth via bearer token for remote access

Docker```dockerfile

FROM node:20-slim
WORKDIR /app
COPY package*.json ./
RUN npm ci --production
COPY dist/ dist/
RUN npm install -g qmd
ENV QMD_BIN=qmd
EXPOSE 3838
CMD ["node", "dist/server.js"]


Mount your memory directory:```bash
docker run -v ~/memory:/memory -e MEMORY_DIR=/memory -p 3838:3838 open-memory

License

MIT

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
open-memory 怎么安装?

在客户端按下面的安装命令接入:node /path/to/open-memory/dist/stdio.js。装完在客户端配置里引用同一个包或端点。

open-memory 怎么在 Claude、Cursor、VS Code 里配置?

它的传输方式是stdio · 本地进程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

open-memory 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

open-memory 需要 API Key 吗?

资料标注需要API Key认证,使用前准备好对应凭据,并按最小权限配置。

open-memory 支持哪些客户端?

资料标注支持 Claude、Cursor、VS Code、Codex;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

open-memory 能提供哪些能力?

资料显示能力标签包括 Local Deployment、Knowledge Base Integration、AI Tool Extension、TypeScript、MCP Protocol、Local Storage。具体工具清单以官方文档为准。

open-memory 需要什么运行环境?

运行时为 node;属于本地进程,需要在本机装好运行环境。

open-memory 是本地还是远程?

它是本地进程(stdio),由客户端在本机拉起。传输方式为stdio · 本地进程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-05-18。