Restora
Read-only MCP server that exposes your local Notion backups to AI agents — query databases, read pages, search a backed-up workspace. Offline by default: no LLM calls, no telemetry, no cloud.
README
Restora MCP Server
A read-only Model Context Protocol server that exposes your local Notion backups to any MCP-capable AI agent — Claude Desktop, Cursor, Cline, Continue, Zed, and others. Ask questions about your backed-up workspace, query databases, and read pages, all from backup files sitting on your own machine.
It's the MCP component of Restora — a Notion backup & restore tool — packaged
here as a standalone, source-available server. The ready-to-run version ships inside
@restora/cli as restora mcp.
Why it's safe
- Offline by default. The server makes zero outbound network calls in its default mode — every
answer is read from local backup
.jsonfiles. The only tool that touches the network isrun_drift_audit, and it is not even registered unless you pass--allow-live. When enabled, it talks only to your own Notion — never to an LLM, never to Restora's servers. - The AI's own model does the reasoning. This server makes no LLM calls, holds no API key, and sends no telemetry.
- Read-only, structurally. Every tool is side-effect-free. Because no tool can write, delete, or take action in Notion, backup content is treated strictly as data, not instructions — a malicious string in a backup can at worst mislead output, never cause an action.
- Path-traversal guarded. A tool's
pathargument must resolve inside your configured backup folder.
Tools
| Tool | What it does |
|---|---|
list_backups |
List your local Restora backup files (path, date, size, database/page counts). |
describe_backup |
Workspace map of a backup: databases → data sources → property schema, the relations graph, and views. Start here. |
query_database |
Rows of one database with readable values — relations resolved to linked page titles. |
get_page |
One page's properties and content, rendered to Markdown or plain text. |
search |
Find pages by case-insensitive substring across titles, property values, and block text. |
read_id_map |
Summarize a restore id-map (restore-map-*.json): old → new ids, to repoint integrations after a restore. |
run_drift_audit |
(requires --allow-live) Compare a backup against your current Notion to find deleted/emptied databases, removed properties, and type changes. Reads your own Notion only. |
Quick start (recommended)
The published package is the fastest path — no clone or build. Add this to your agent's MCP config:
{
"mcpServers": {
"restora": {
"command": "npx",
"args": ["-y", "@restora/cli", "mcp"],
"env": {}
}
}
}
Config file locations:
- Claude Desktop (macOS):
~/Library/Application Support/Claude/claude_desktop_config.json - Claude Desktop (Windows):
%APPDATA%\Claude\claude_desktop_config.json - Cursor:
~/.cursor/mcp.json(or.cursor/mcp.jsonin a project)
To enable the live drift-audit tool, add "--allow-live" to args. Run npx @restora/cli mcp --print-config
to print a ready-to-paste config.
Build from source (this repo)
npm install
npm run build # → dist/restora-mcp.js (single self-contained bundle, zero runtime deps)
node dist/restora-mcp.js --print-config
Point an agent at the built binary:
{
"mcpServers": {
"restora": { "command": "node", "args": ["/absolute/path/to/dist/restora-mcp.js"], "env": {} }
}
}
Flags: --file <path> (a specific backup), --dir <path> (a backups folder), --allow-live (enable
run_drift_audit), --print-config (print an MCP config and exit).
Getting backups to read
This server reads Restora backup files. Create some for free with the CLI:
npx @restora/cli backup --to local
or from restora.cc. Backups are plain .json — your data stays on your machine.
License
MIT © Taha Bakri
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。