remarkable-mcp
Gives AI assistants direct control of a reMarkable tablet over SSH, enabling listing notes, uploading PDFs, exporting handwriting, and writing on pages via a single MCP tool.
README
remarkable-mcp
Give an AI assistant direct control of your reMarkable tablet — list notes, upload PDFs, export handwriting, write on pages — over SSH.
The assistant gets one MCP tool, remarkable_execute, and writes a short JavaScript snippet that calls rme.* (list, mkdir, write, export, …) in one shot.
| You have | This project is |
|---|---|
| A reMarkable with developer mode | An MCP server |
SSH over USB (10.11.99.1) or Wi‑Fi |
stdio or Streamable HTTP (Docker) |
| Claude / Cursor / VS Code / any MCP client | SFTP into the tablet’s xochitl library |
How it works
MCP client this process tablet
─────────── ──────────── ──────
remarkable_execute({ node:vm sandbox SSH/SFTP
code: `async () => { → rme.list / rme.mkdir / … → /home/root/.local/share/remarkable/xochitl
await rme.mkdir(…)
return await rme.list({})
}`
})
- Enable Settings → General → Developer mode on the tablet.
- SSH in over USB (
root@10.11.99.1) or Wi‑Fi (IP under About → Copyrights). - Run this server. Point your MCP client at it.
Install
Node 22+. No clone required.
Claude Desktop, Cursor, VS Code, …
{
"mcpServers": {
"remarkable": {
"command": "npx",
"args": ["-y", "@lukaisailovic/remarkable-mcp"],
"env": {
"REMARKABLE_HOST": "10.11.99.1",
"REMARKABLE_USER": "root",
"REMARKABLE_PASSWORD": "your-tablet-password"
}
}
}
}
Prefer a key? Set REMARKABLE_KEY to a private-key path (or the PEM). If unset, ~/.ssh/id_ed25519 then id_rsa are tried.
HTTP
Same server, Streamable HTTP on /mcp:
npx -y @lukaisailovic/remarkable-mcp --http
# → http://127.0.0.1:8080/mcp health: /health
Or pull the image from GHCR:
docker run --rm -p 8080:8080 \
-e REMARKABLE_HOST=10.11.99.1 \
-e REMARKABLE_PASSWORD=your-tablet-password \
ghcr.io/lukaisailovic/remarkable-mcp:latest
From a clone: docker compose up --build (bind 0.0.0.0:8080, pass REMARKABLE_*).
A local checkout’s .mcp.json watches src/ and restarts the stdio child on save (no client restart). pnpm dev:http does the same for Streamable HTTP. In Grok, /mcps disable+enable respawns the process; r only refreshes the list.
What the model can do
The client only lists remarkable_execute. Inside the snippet, methods live on rme.
async () => {
const folder = await rme.mkdir({ name: "Projects" });
await rme.createNotebook({ name: "Ideas", parent: folder.id });
await rme.writeText({
notebook: "Ideas",
newPage: true,
blocks: [
{ text: "Ship it", style: "title" },
{ text: "Talk to design", style: "checkbox" },
],
});
await rme.writeText({ notebook: "Ideas", text: "Write the RFC", style: "checkbox" });
return await rme.read({ notebook: "Ideas", page: 2 });
};
A notebook is the file (name, path /Work/Notes, or UUID). A page is 1-based inside that notebook. writeText always writes native Type Folio text and appends — call it again to stack more paragraphs. Ink points are [x, y] in 0–1 from the top-left. remove moves to trash. Writes restart xochitl once after the tool returns so the UI refreshes.
rme.* |
What it does |
|---|---|
list({ includeTrash?, folder? }) |
Library listing (trash hidden by default) |
browse({ path? }) |
One folder, or a single notebook |
search({ query, tag? }) |
Name / path search, optional tag |
info({ notebook }) |
Id, path, type, tags, page count, pages[].title |
read({ notebook, page? }) |
Native paragraphs (and checkbox state), or all pages if page is omitted. PDF/EPUB text. |
download({ notebook }) |
Raw PDF/EPUB as base64 |
exportPage({ notebook, page?, format? }) |
Ink on a page → png or svg (typed text is not drawn) |
upload({ name, dataBase64, parent?, fileType? }) |
Put a PDF or EPUB on the tablet |
mkdir / move / rename / remove |
Folders and trash |
createNotebook / addPage / removePage |
Notebooks and pages |
writeInk({ notebook, strokes, page? }) |
Pen / highlighter strokes |
writeMermaid({ notebook, mermaid, page? }) |
Mermaid → ink (flowchart, sequence, state, class, ER, xychart) |
writeText({ notebook, text?, style?, checked?, blocks?, page?, newPage?, replace? }) |
Native Type Folio: title heading body bullet checkbox. Stacks unless replace. |
tag({ notebook, tag, remove?, page? }) / tags() |
Notebook or page tags |
refresh() |
Restart xochitl |
Configuration
Flags and env vars are interchangeable (--host ≡ REMARKABLE_HOST).
| Tablet | Default | |
|---|---|---|
REMARKABLE_HOST / --host |
10.11.99.1 |
SSH host (USB default) |
REMARKABLE_USER / --user |
root |
SSH user |
REMARKABLE_PORT / --port |
22 |
SSH port |
REMARKABLE_PASSWORD / --password |
Password | |
REMARKABLE_KEY / --key |
~/.ssh/id_ed25519, then id_rsa |
Key path or PEM |
| Server | Default | |
|---|---|---|
--http / MCP_HTTP=1 |
off | Streamable HTTP instead of stdio |
MCP_HTTP_HOST / --http-host |
127.0.0.1 |
Bind address (0.0.0.0 in Docker) |
MCP_HTTP_PORT / PORT / --http-port |
8080 |
HTTP port |
--fake / REMARKABLE_FAKE=1 |
off | In-memory tablet (tests, no device) |
--fake-dir / REMARKABLE_FAKE_DIR |
Local directory treated as a xochitl tree |
Agent skill
Installable with skills.sh:
npx skills add lukaisailovic/remarkable-mcp
The skill lives at skills/remarkable-mcp/SKILL.md.
Develop
pnpm test
pnpm lint
pnpm fmt
pnpm build
pnpm exec tsx src/index.ts --fake
pnpm exec tsx src/index.ts --fake --http --http-port 8080
Tests use a fake tablet filesystem. A live device is never required.
Inspired by sammorrowdrums/remarkable-mcp and itsfabioroma/remarkable-cli.
MIT. See LICENSE.
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。