remarkable-mcp 0.3.0

MCP server for accessing reMarkable tablet data

reMarkable MCP Server

An MCP (Model Context Protocol) server for accessing your reMarkable tablet data through the reMarkable Cloud.

Quick Install

What It Does

• Read typed text directly from notebooks (v3+ software, no OCR needed)
• Browse and search your document library
• Access recent files with content previews
• OCR for handwritten content (Google Vision API recommended)
• MCP resources and prompts for deeper integration

⚡ Recommended: SSH Mode + Google Vision

For the best experience, we strongly recommend:

1. SSH Mode — 10-100x faster than Cloud API, works offline, no subscription needed
2. Google Vision API — Far superior handwriting recognition compared to Tesseract

See SSH Mode Setup and OCR Configuration below.

Installation

Using uvx (Recommended)

# Get your reMarkable token
uvx remarkable-mcp --register YOUR_ONE_TIME_CODE

Click the Quick Install badges above, or configure manually.

From Source

git clone https://github.com/SamMorrowDrums/remarkable-mcp.git
cd remarkable-mcp
uv sync
uv run python server.py --register YOUR_ONE_TIME_CODE

Setup

SSH Mode (Recommended)

Connect directly to your reMarkable via USB — 10-100x faster than Cloud API, works offline, and doesn't require a Connect subscription.

📖 Detailed SSH setup guide: remarkable.guide/guide/access/ssh.html

Requirements

1. Developer Mode enabled on your reMarkable tablet

   • Go to Settings → General → Software → Developer mode
   • This is required even if you have a Connect subscription
   • ⚠️ Enabling developer mode will factory reset your device (back up first!)

2. USB connection to your computer

   • Connect via the USB-C cable
   • Your tablet must be on and unlocked

3. SSH access (automatic with developer mode)

   • Default IP over USB: 10.11.99.1
   • Password shown in Settings → General → Software → Developer mode

Configure MCP for SSH

VS Code — Add to .vscode/mcp.json:

{
  "servers": {
    "remarkable": {
      "command": "uvx",
      "args": ["remarkable-mcp", "--ssh"],
      "env": {
        "GOOGLE_VISION_API_KEY": "your-api-key"
      }
    }
  }
}

That's it! Default connection is root@10.11.99.1 (standard USB IP).

Custom SSH Host

Set up passwordless SSH for convenience:

# Copy your SSH key to the tablet
ssh-copy-id root@10.11.99.1

# Or add to ~/.ssh/config:
Host remarkable
    HostName 10.11.99.1
    User root

Then use the alias:

{
  "servers": {
    "remarkable": {
      "command": "uvx",
      "args": ["remarkable-mcp", "--ssh"],
      "env": {
        "REMARKABLE_SSH_HOST": "remarkable",
        "GOOGLE_VISION_API_KEY": "your-api-key"
      }
    }
  }
}

SSH Environment Variables

Variable	Default	Description
REMARKABLE_SSH_HOST	10.11.99.1	SSH hostname or IP
REMARKABLE_SSH_USER	root	SSH username
REMARKABLE_SSH_PORT	22	SSH port

Cloud API (Alternative)

If you can't enable developer mode, you can use the Cloud API. This is slower and requires a reMarkable Connect subscription.

1. Get a One-Time Code

Go to my.remarkable.com/device/desktop/connect and generate a code.

2. Convert to Token

uvx remarkable-mcp --register YOUR_CODE

3. Configure MCP

VS Code — Add to .vscode/mcp.json:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "remarkable-token",
      "description": "reMarkable API Token",
      "password": true
    }
  ],
  "servers": {
    "remarkable": {
      "command": "uvx",
      "args": ["remarkable-mcp"],
      "env": {
        "REMARKABLE_TOKEN": "${input:remarkable-token}",
        "GOOGLE_VISION_API_KEY": "your-api-key"
      }
    }
  }
}

Your token is stored securely using VS Code's input system with password: true.

Claude Desktop — Add to claude_desktop_config.json:

{
  "mcpServers": {
    "remarkable": {
      "command": "uvx",
      "args": ["remarkable-mcp"],
      "env": {
        "REMARKABLE_TOKEN": "your-token-from-step-2",
        "GOOGLE_VISION_API_KEY": "your-api-key"
      }
    }
  }
}

Tools

Tool	Description
remarkable_read	Extract text from a document (with pagination and search)
remarkable_browse	List files or search by name
remarkable_recent	Get recently modified documents
remarkable_status	Check connection status

All tools are read-only and return structured JSON with hints for next actions.

remarkable_read Parameters

Parameter	Type	Default	Description
document	string	required	Document name or path
content_type	string	"all"	"all", "annotations", or "raw"
page	int	1	Page number for pagination
grep	string	None	Search for keywords (returns matches with context)
include_ocr	bool	False	Enable OCR for handwritten content

Content Types:

• "all" — Raw document text plus annotations/highlights (default)
• "annotations" — Only typed text, highlights, and OCR content from notebooks
• "raw" — Only raw PDF/EPUB text (no annotations)

Pagination:

• PDF/EPUB: Pages are ~8000 character chunks of text
• Notebooks with OCR: Pages correspond to actual notebook pages

Pagination Output:

{
  "content": "...",
  "page": 1,
  "total_pages": 42,
  "more": true,
  "next_page": 2
}

Grep Output:

{
  "content": "...context around matches...",
  "grep_term": "search term",
  "grep_matches": 15,
  "page": 1
}

Resources

Documents are automatically registered as MCP resources on startup.

URI Scheme	Description
remarkable:///{path}.txt	Extracted text content from any document
remarkableraw:///{path}.pdf	Raw PDF file (SSH mode only)
remarkableraw:///{path}.epub	Raw EPUB file (SSH mode only)

Text Resources (remarkable:///)

Each document is registered with its full path. Returns extracted text content:

• PDF/EPUB: Full text content extracted from the source file
• Notebooks: Typed text (Type Folio), highlights, and annotations
• Handwritten content via OCR (if enabled)

Raw Resources (remarkableraw:///)

PDF and EPUB files are also registered as raw resources in SSH mode. Returns the original file as base64-encoded data. Cloud API doesn't provide access to source files, so raw resources are only available in SSH mode.

Prompts

summarize_recent · find_notes · daily_review · export_document · organize_library · meeting_notes

Usage

# Read first page of a document
remarkable_read("Meeting Notes - Nov 2025")

# Read page 3 of a long document
remarkable_read("My Book.epub", page=3)

# Search for keywords in a document
remarkable_read("Project Plan", grep="deadline")

# Only get annotations (typed text, highlights, OCR)
remarkable_read("November journal 2025", content_type="annotations", include_ocr=True)

# Only get raw PDF/EPUB text (no annotations)
remarkable_read("Research Paper.pdf", content_type="raw")

# Browse and search
remarkable_browse("/")
remarkable_browse(query="meeting")
remarkable_recent(limit=5, include_preview=True)

Text Extraction

PDF and EPUB Documents

Text is extracted directly from PDF and EPUB files using PyMuPDF and ebooklib. This provides the full document content without needing OCR.

Notebooks

Typed text from v3+ notebooks is extracted natively via rmscene — no OCR required.

Handwritten content uses OCR. Two backends are supported:

Google Cloud Vision (Strongly Recommended)

Google Vision is far superior to Tesseract for handwriting recognition. Unless your handwriting is exceptionally clear and print-like, Tesseract will produce mostly gibberish. Google Vision handles cursive, messy handwriting, and mixed text/drawings much better.

Quick Setup with API Key (Easiest)

1. Go to Google Cloud Console
2. Create a project (or select existing)
3. Enable the Cloud Vision API
4. Go to Credentials → Create Credentials → API Key
5. Add the key to your MCP config:

{
  "env": {
    "GOOGLE_VISION_API_KEY": "your-api-key"
  }
}

Cost: Vision API offers 1,000 free requests/month. After that, ~$1.50 per 1,000 images.

Alternative: Service Account Credentials

For production use or tighter security:

# Set up credentials (one of these methods):
# 1. Set GOOGLE_APPLICATION_CREDENTIALS to your service account JSON file
# 2. Run `gcloud auth application-default login` for development
# 3. Use a GCP environment with default credentials (Cloud Run, GKE, etc.)

# Install the optional SDK dependency
pip install remarkable-mcp[ocr]

Tesseract (Fallback — Not Recommended for Handwriting)

Tesseract is designed for printed text, not handwriting. It will be used as a fallback if Google Vision is not configured, but expect poor results on handwritten notes.

Only use Tesseract if:

• You have very clear, print-like handwriting
• You need fully offline OCR
• You're only processing printed documents

# macOS
brew install tesseract

# Ubuntu/Debian
sudo apt-get install tesseract-ocr

# Arch
sudo pacman -S tesseract

OCR Configuration

Environment Variable	Values	Description
GOOGLE_VISION_API_KEY	API key string	Google Vision API key (recommended)
REMARKABLE_OCR_BACKEND	auto, google, tesseract	Force a specific backend. Default: auto (uses Google if configured)

PDF highlights and annotations are also extracted automatically.

Performance: OCR results are cached per document, so reading multiple pages of the same notebook doesn't re-run OCR.

Behavior Notes

• Archived documents (synced=false or in trash) are automatically filtered from browse, search, and recent results
• Case-insensitive matching for document names and paths
• Page ordering for notebooks uses the actual page order from the device

Design

Intent-based tools that map to what you actually want to do. Responses include hints for logical next steps. Errors explain what went wrong and how to fix it. Four tools cover most use cases.

Authentication

Set REMARKABLE_TOKEN in your MCP config (recommended), or the server will fall back to ~/.rmapi.

Development

uv sync --all-extras
uv run pytest test_server.py -v
uv run ruff check .
uv run ruff format .

Project Structure

remarkable-mcp/
├── server.py           # Entry point
├── remarkable_mcp/     # Main package
│   ├── server.py       # FastMCP server
│   ├── api.py          # Cloud API helpers
│   ├── extract.py      # Text extraction
│   ├── tools.py        # MCP tools
│   ├── resources.py    # MCP resources
│   └── prompts.py      # MCP prompts
└── test_server.py      # Tests

License

MIT

---

Built with rmscene, inspiration from ddvk/rmapi for the sync protocol, and Scrybble for text extraction ideas.