rmink

Turn images, photos of handwriting, SVGs, logos and PDFs into native, editable reMarkable ink.

Most tools put a picture behind your notes. rmink writes real pen strokes (.rm v6, the format of firmware 3.x) into a .rmdoc notebook, so on the tablet you can erase part of a converted sketch, lasso a word and move it, or recolour a diagram.

Supported devices: reMarkable 2, Paper Pure, Paper Pro (colour) and Paper Pro Move (colour).

Install

pip install rmink            # or run without installing: uvx rmink ...
pip install 'rmink[mcp]'     # + MCP server for Claude and other agents
pip install 'rmink[heic]'    # + iPhone HEIC photos

Every command also works without installing anything: uvx rmink convert notes.jpg (needs uv).

Use

rmink convert notes.jpg                          # -> notes.rmdoc for reMarkable 2
rmink convert logo.png -d pro --preview logo.png # colour, plus a PNG preview
rmink convert paper.pdf --pages 1-3 --fit width -o paper.rmdoc
rmink convert a.jpg b.jpg c.pdf --name "Week 41" # several inputs -> one notebook
rmink preview sketch.jpg -o preview.png          # preview only, no notebook
rmink inspect paper.rmdoc                        # pages, strokes, pens, colours
rmink repair broken.rmdoc.zip                    # fix a notebook a Mac unpacked and re-zipped
rmink devices                                    # device profiles
rmink calibrate -d pro                           # ruler page for checking scale

Useful options: --mode auto|vector|centerline|outline, --colors auto|mono|N, --pen fineliner|ballpoint|marker|pencil, --width <px>, --fill hatch|outline|none, --threshold auto|0-255, --simplify <px>, --max-points <n>.

Shading. Photos and portraits can keep their grey tones: --shading stipple (dots, denser where darker), halftone (a dot screen with growing dots), hatch (lines, closer where darker), crosshatch (more line directions where darker) or pencil (pencil strokes whose pressure follows the tone). Tune with --shade-density and --shade-contrast. The web UI shows a small preview of every style side by side, so you can pick one before you download. Run rmink convert --help for the full list.

Web UI

pip install 'rmink[web]'
rmink web                     # opens http://localhost:8765

Drop files on the page, choose a device and options, and the preview updates as you change them. When it looks right, download the .rmdoc.

To host it for others, deploy deploy/vercel/ on Vercel (see hosting), or use the included Dockerfile (docker build -t rmink . && docker run -p 8765:8765 rmink). The per-request limits can be tightened with RMINK_MAX_UPLOAD_MB, RMINK_MAX_FILES, RMINK_MAX_PAGES, RMINK_TIMEOUT_S and RMINK_CONCURRENCY; see src/rmink/web/__init__.py. Uploads are deleted as soon as the request finishes.

Getting it onto the tablet: connect the tablet by USB, turn on Settings → Storage → USB web interface, open http://10.11.99.1 and drag the .rmdoc in (100 MB limit). Alternatively, drag it into the reMarkable desktop app.

Or send it through the reMarkable Cloud: if rmapi is installed (or RMINK_RMAPI points at it), the web UI shows a Send to reMarkable button. The first time, it asks for a one-time code from https://my.remarkable.com/device/browser/connect; the connection is remembered in your browser. Notebooks land in the root of My files.

If a Mac turned the download into a folder or a .zip and the tablet will not import it, run rmink repair on it, or use Repair a notebook in the web UI.

Use with Claude

MCP server (Claude Code, Claude Desktop, any MCP client):

claude mcp add rmink -- uvx --from 'rmink[mcp]' rmink mcp

Tools:

• preview_ink returns a PNG, so the model can look at the result and tune the options before converting.
• convert_to_rmdoc writes the notebook.
• inspect_rmdoc, list_devices and calibration_sheet.

Skill: copy skills/rmink/ into ~/.claude/skills/. It covers choosing a mode, the preview → tune → convert loop, and transferring the file.

Workbooks

Claude writes a workbook on a topic: a PDF in the tablet's page format with a linked table of contents, short sections and ruled answer boxes. You write the answers by hand. The next round reads each box, has Claude transcribe it word for word, grades it against Claude's solution, and sends a new revision with the answers set as PDF text and a results page at the end. The handwriting is either removed or kept on a hidden layer. The solutions are on a hidden "Facit" layer, from the start in practice mode or after grading in a test. Questions can also be multiple choice ("choose one" or "choose all that apply"). rmink reads and grades the crosses itself. In practice mode a wrong answer gets a hint and a new attempt, up to three in all.

rmink workbook new "Fractions for year 5" --name Fractions --mode test   # needs rmink[rounds]
rmink workbook round Fractions --ink keep

The MCP tools are create_workbook, read_workbook and save_workbook_answers. The owner's /rounds page can do the same. Details are in docs/workbook.md (Swedish).

How it works

input ─► ingest ─┬─ vector (PDF→SVG with text as paths, SVG) ─► paths → strokes, fills → outline + hatch
                 └─ raster (photos, scans, images)
                       background normalisation + Otsu/Sauvola → ink mask
                       colour clustering (CIELAB k-means) → device palette
                       per component: thin → skeleton graph → centerline strokes (+ width)
                                      thick → contours + hatching
                       optional shading: tone map → dots / dot screen / hatching / pencil pressure
        ─► simplify (RDP), smooth (Chaikin), chain, reading order, point budget
        ─► .rm v6 via rmscene ─► .rmdoc (content/metadata/thumbnails)

Diagrams of the whole app and the building knowledge base (in Swedish, built from the code) are in docs/diagrams/.

Code layout: src/rmink/ contains ingest/, extract/, geometry/, color/, writer/rm_v6.py, package/rmdoc.py, render/preview.py, api.py, cli.py, mcp_server.py and web/ (FastAPI plus a single static page).

Device notes

The reMarkable 2 coordinate system is well known: 1404×1872, with x measured from the page centre. The other profiles assume the same scene unit on a larger canvas, and are marked beta until that is confirmed on hardware. Print rmink calibrate -d <device> and measure it; if it is off, override the profile in ~/.config/rmink/devices.toml:

[pro]
ink_scale = 0.99

See docs/device-acceptance.md for the on-device checklist.

Development

python -m venv .venv && .venv/bin/pip install -e '.[dev]'
.venv/bin/pytest            # unit, round-trip, fidelity, CLI and MCP tests
python examples/make_examples.py

Releases: see docs/releasing.md.

The fidelity tests render the parsed output and compare it with the source ink, checking precision and recall within 3 px. That way the tests cover the writer as well as the tracing.

Support

rmink is free, and that is not going to change. If it saves you time and you want to say thanks, you can support it with 100 kr via Stripe. The money pays for the hosted web service and further development. Starring the repo and reporting how it works on your tablet in an issue helps a lot too.

License

rmink is free software under the GNU AGPL-3.0 or later. This is required by PyMuPDF, which is AGPL. If you host a modified version as a web service, you have to offer its source code to the people who use it. Hosted instances set RMINK_SOURCE_URL so the footer links to the source.

Not affiliated with or endorsed by reMarkable AS. "reMarkable" is a trademark of reMarkable AS.

History

This project began as pdf2rmnotebook by Simon Rechermann, a wrapper around drawj2d that wrote v5 pages in black fineliner only. That version lives on at the tag v1.1.0. rmink 2.0 replaces it with a pure-Python pipeline: no Java, v6 output, colour, image input, centerline tracing and an agent interface.