mcp-memory-vault
A self-hosted memory layer for Claude using plain markdown files, enabling persistent context across different Claude accounts.
README
📦 Memory Vault — Persistent Context Layer for Claude (MCP)
Memory Vault is a self-hosted, local-first memory engine for Claude. Your context lives as human-readable Markdown files in an Obsidian-compatible vault on your machine, or you can back it with MongoDB Atlas for cloud persistence. The goal is simple: keep your engineering context portable across accounts, devices, and client surfaces without locking it to a single AI vendor.
Built on Anthropic's Model Context Protocol (MCP).
Features
- Zero account lock-in: keep the same context across personal or corporate Claude accounts.
- Obsidian native: memories are plain Markdown files, so they show up directly in Obsidian.
- MongoDB Atlas integration: optionally sync or back up memory data in the cloud.
- Dual transports: run locally over
stdiofor Claude Desktop or over HTTP for broader access. - Privacy first: no tracking telemetry; you own the vault and any cloud keys you configure.
Architecture
flowchart TB
client[Claude Desktop / Browser Client] -->|MCP Protocol| server[Memory Vault Server (Node)]
server -->|Local file I/O| vault[Obsidian Vault\n(Plain Markdown Notes)]
server -->|Cloud sync| atlas[MongoDB Atlas\n(Persistent datastore)]
Quick Start
git clone https://github.com/YOUR_USERNAME/mcp-memory-vault.git
cd mcp-memory-vault
npm install
npm link
memory-vault-setup
The setup wizard detects your operating system, locates the local Claude Desktop configuration, and injects the MCP server entry for you.
Configuration
Copy the template file and edit the values for your environment:
cp .env.example .env
Example configuration:
VAULT_DIR=/path/to/your/Obsidian/Vault/claude-memory
PORT=8080
MCP_SHARED_SECRET=your-secure-random-string
MONGODB_URI=mongodb+srv://<username>:<password>@cluster.mongodb.net/memory_vault?retryWrites=true&w=majority
Use VAULT_DIR for the local Markdown vault, PORT for the HTTP server, MCP_SHARED_SECRET for connector access, and MONGODB_URI if you want MongoDB-backed storage.
Tool Surface
The server exposes five MCP tools:
| Tool | Parameters | Purpose |
|---|---|---|
save_memory |
topic, content, tags |
Save a memory or decision into the persistent vault. |
get_memory |
topic |
Retrieve everything stored under a topic. |
search_memory |
query, topic, limit |
Search memories with optional topic scoping. |
list_memory_topics |
None | List all stored topics and their metadata. |
delete_memory_topic |
topic |
Delete an entire topic from the vault. |
Claude Desktop Setup
The included setup script writes the Claude Desktop MCP configuration for you.
memory-vault-setup
If you prefer to configure it manually, the server entry points to src/stdio-server.js and passes VAULT_DIR through the environment.
Hosted HTTP Deployment
To expose the vault over HTTP for Claude web or other remote clients:
- Push the repository to GitHub.
- Deploy it as a web service on Render.
- Use
npm installas the build command andnpm run start:serveras the start command. - Set
MCP_SHARED_SECRETandMONGODB_URIin the Render environment settings. - Add a custom connector in Claude and point it at your deployment URL.
Example connector URL:
https://your-app.onrender.com/mcp?secret=YOUR_MCP_SHARED_SECRET_HERE
License
MIT © 2026 Akshat
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。