Reported issues for reMarkable MCP Server by SamMorrowDrums
Pod holds 22 of 45 GitHub reports that passed its relevance review. This can include external user reports, maintainer-confirmed bugs, and concrete feature gaps. Treat them as evidence to inspect, not a count of distinct defects.
Back to reMarkable MCP Server by SamMorrowDrums.
Most discussed
Can't get it to work
Man, I love the idea of this project, but I haven't been able to get it to work. In USB mode it appears that the image generation is failing, so it can't do OCR to extract text. Even if cloud mode is slow, I'd be willing to live with it to give it a go, but it fails with bad request, despite me copying and pasting the generated key output from the terminal.
Any suggestions are welcome!
Read the thread · 2026-05-20 · closed · external user · 6 comments
remarkable_image fails in Claude Desktop/Web with "cairo library missing", but works in MCP Inspector
I'm seeing inconsistent behavior between MCP Inspector and Claude.
remarkable_browseworks correctly.remarkable_readworks correctly.remarkable_imageworks correctly in MCP Inspector.- However, Claude always reports "cairo library missing" when it tries to use
remarkable_image.
Because Inspector successfully generates PNG images, it doesn't appear that Cairo is actually missing on the server.
Environment
- remarkable-mcp: 1.0.0
- Started with:
npx -y mcp-
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/137) · 2026-07-21 · closed · external user · 4 comments
### 400 Bad Request on sync/v3 endpoint - Cloud authentication fails despite valid device token
**Environment**
- OS: Windows 11
- remarkable-mcp version: latest (uvx remarkable-mcp)
- Transport: Cloud (REMARKABLE_TOKEN via environment variable)
- reMarkable Connect subscription: active
**Problem**
Authentication always fails with 400 Bad Request on a specific sync/v3 URL.
The URL and hash are always identical regardless of token:
400 Client Error: Bad Request for url:
https://internal.cloud.remarkable.com/sync/v3/files/8a2469797b082cbe6ad8ff9fc5a4b8ca161cc44f59542ebc9d3ace1ce5ee32dc
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/101) · 2026-06-06 · closed · external user · 4 comments
### Export in pdf, ocr via google gemini
Thanks for your incredible work!
I'm wondering why you chose google vision api, if let's say gemini flash can do the trick with pdf even simpplier and with a way better quality?
Thanks, af
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/44) · 2026-01-12 · closed · external user · 4 comments
### Add a pluggable OCR provider interface
## Summary
Add more OCR provider options beyond Google Vision and Tesseract. Goal: **BYOK (Bring Your Own Key)** — let users choose their preferred provider.
## Current State
| Provider | Status | Notes |
|----------|--------|-------|
| Google Vision | ✅ Implemented | Excellent handwriting recognition |
| Tesseract | ✅ Implemented | Offline fallback, poor for handwriting |
## Proposed Additions
| Provider | Priority | Notes |
|----------|----------|-------|
| **Microsoft Azure AI Vision** |
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/25) · 2025-11-28 · open · 4 comments
### Mark notebook pages as processed
## Summary
Add write capabilities to remarkable-mcp, enabling document creation and synchronization.
## Proposed Features
### Phase 1: Basic Write Operations
- **Create notebooks** — Create new blank notebooks in any folder
- **Upload PDFs** — Upload PDF files to reMarkable
- **Folder management** — Create, rename, and move folders
### Phase 2: Obsidian Integration
- **Markdown to PDF** — Convert markdown notes to PDFs and upload
- **Bi-directional sync** — Notes flow between reMarkable and
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/24) · 2025-11-28 · open · 4 comments
### SSH smoke sessions intermittently lose connectivity on large libraries
## Summary
On an up-to-date reMarkable Paper Pro with roughly 400 synced documents, a sequence of SSH write tools can still leave the SSH daemon accepting TCP connections but unable to complete the SSH banner exchange. This causes subsequent authoring and cleanup operations to fail even after #131's xochitl readiness/settle changes.
## Environment
- Device: reMarkable Paper Pro, current firmware
- Transport: USB SSH at `10.11.99.1`
- Library: approximately 399–404 documents
- remarkable-mcp:
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/157) · 2026-08-14 · closed · 3 comments
### remarkable_read fails in cloud mode with coroutine error
## Description
`remarkable_read()` fails in cloud mode (both 0.7.0 and 0.8.0) with the following error:
```json
{
"_error": {
"type": "read_failed",
"message": "the JSON object must be str, bytes or bytearray, not coroutine",
"suggestion": "Check remarkable_status() to verify your connection."
}
}
Reproduction
- Connect via cloud mode (environment variable token)
- Verify connection works:
remarkable_status()→authenticated: true,document_count: 67 - Verify
Read the thread · 2026-03-05 · closed · external user · 3 comments
Most recent
Avoid event-loop blocking after USB metadata cache invalidation race
Problem
USBWebClient.run_method_async() checks whether get_meta_items is cached, releases the metadata lock, then may call get_meta_items() synchronously. If another thread invalidates the cache in that narrow gap, the second check starts a full USB HTTP traversal on the asyncio event-loop thread.
Impact
Results remain correct, but all MCP tool handling can pause for the duration of the metadata traversal.
Expected behavior
Keep the cached fast path, but ensure a cache miss a
Read the thread · 2026-08-17 · open · 0 comments
Exclude trashed documents from lookup and resource registration
Problem
A trashed document keeps its visible name. If a live document has the same name, by-name resolution can select the trashed item first. Trashed documents may also be registered as MCP resources.
Reported by @lowercasename in #163.
Scope
Use the existing _is_cloud_archived transport-aware predicate rather than adding another trash helper. Apply it consistently to:
- shared
remarkable_read/remarkable_imagedocument lookup remarkable_canvaslookup- synchronous resourc
Read the thread · 2026-08-14 · closed · 1 comment
Protect USB web document caches during concurrent dispatch
Problem
USBWebClient permits concurrent operations but reads and replaces _documents and _documents_by_id without a shared lock. A concurrent reader can observe the updated list with the previous ID map, or vice versa.
Impact
The failure is transient and does not change tablet data, but it can produce stale or inconsistent document resolution under concurrent USB operations.
Expected behavior
Use one metadata lock and a generation-aware or atomic cache replacement pattern, ma
Read the thread · 2026-08-14 · closed · 0 comments
Refresh after cancellation when an SSH write may have started
Problem
If OperationDispatcher.close() cancels an active SSH write after the remote-execution marker was emitted, _run_process currently raises SSHJobCancelled. SSHRefreshCoordinator.run_write treats that as a clean cancellation and does not schedule a safety refresh, even though the remote mutation may already have started.
Impact
Local caches are invalidated and recover on the next read, but the tablet UI may not notice the filesystem mutation until another write or a manual `
Read the thread · 2026-08-14 · closed · 0 comments
remarkable_read and remarkable_image disagree on total_pages for handwritten notebooks
One more from Claude, so the usual grain of salt applies:
For a notebook that's mostly handwriting, remarkable_read reports total_pages as the number of extracted text segments (len(notebook_pages)), which collapses to 1 when there's little or no typed text. remarkable_image reports the real document page count. So the same seven page notebook answers 1 from one tool and 7 from the other.
This affects any client that reads first and then iterates pages using the count it was given. Not
Read the thread · 2026-08-14 · closed · external user · 1 comment
render_merged is opt-in, so agents silently get inkless renders of annotated PDFs
Claude has filed this one too, but pretty sure it's right!
After #159 the merged PDF and annotation rendering works really well, but remarkable_image defaults to render_merged=False. A client that doesn't pass the flag follows the annotation-only path: a page with ink renders as bare strokes floating on the background with no PDF page beneath them, and a page without ink falls back to rasterizing the clean source PDF. Neither response mentions the missing layer. This is a bad bug for agent
Read the thread · 2026-08-14 · closed · external user · 1 comment
default nix configuration module does not even build
git commit version: d0f98094626bb2fb49c32b10bfb8fcc6b732966d
error:
building the system configuration...
error:
… while calling the 'derivationStrict' builtin
at «nix-internal»/derivation-internal.nix:37:12:
36|
37| strict = derivationStrict drvAttrs;
| ^
38|
… while evaluating derivation 'nixos-system-nixos-26.11pre1050399.279b4a8275f0'
whose name attribute is located at /nix/store/dy2m3gv51jdklsi
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/162) · 2026-08-14 · closed · external user · 1 comment
### Synchronize reMarkable documents with an Obsidian vault
## Problem
Users want reMarkable notes to flow into an Obsidian vault without manually exporting, naming, and placing each document.
## Scope
Design an opt-in synchronization workflow that can:
- export selected notebooks or folders into an Obsidian vault
- preserve stable document identity so repeated runs update rather than duplicate notes
- map reMarkable folders and useful metadata into predictable Markdown paths/frontmatter
- detect conflicts and avoid silently overwriting user-edited O
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/158) · 2026-08-14 · open · 0 comments
### Fresh installs broken: unbounded mcp dependency now resolves to the 2.0 SDK
Thanks for building this! Browsing and OCR-ing a reMarkable library over MCP is exactly what I needed, and the transport fallback design is really nice. This is going to make using my tablet so much nicer!
Claude has filed this, so please take with a grain of salt:
Fresh installs are currently broken: `pyproject.toml` declares `mcp>=1.27.0` with no upper bound, and the MCP Python SDK has released 2.0.0, which removes the `mcp.server.fastmcp` module path that `server.py` imports. A clean instal
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/149) · 2026-08-11 · closed · external user · 1 comment
### ModuleNotFoundError: No module named 'mcp.server.fastmcp' when installed via uvx (unpinned mcp dependency)
## Description
Running `uvx remarkable-mcp --usb` (or `--ssh`) fails at startup with:
## Root cause (suspected)
`remarkable-mcp` does not appear to pin an upper bound on the `mcp` SDK
dependency. `uvx` resolves the latest available `mcp` release at each
invocation, and a recent `mcp` release appears to have moved or removed the
`mcp.server.fastmcp` module, breaking the import in `server.py`.
## Reproduction
- Worked correctly on 2026-07-28 (fresh `uvx` install, USB mode).
- Broke starting 2
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/143) · 2026-08-03 · closed · external user · 0 comments
### Unbounded mcp>=1.27.0 breaks on mcp 2.0.0 — server dies at import as a generic "Connection closed"
## Summary
`remarkable-mcp` declares `mcp>=1.27.0` with no upper bound. `mcp` 2.0.0 removed the `mcp.server.fastmcp` module, so any fresh resolve picks up `mcp>=2` and the server dies at import.
Because `uvx` re-resolves on every launch, this breaks the server *silently and repeatedly* on a previously working install — no code changed on either side.
## What the user sees
The MCP client reports:
-32000: Connection closed
That reads like an authentication or cloud-token failure, whi
[Read the thread](https://github.com/SamMorrowDrums/remarkable-mcp/issues/141) · 2026-07-30 · closed · external user · 0 comments
### ModuleNotFoundError: No module named 'mcp.server.fastmcp'
Got this error today:
Traceback (most recent call last):
File "
Read the thread · 2026-07-28 · closed · external user · 1 comment
Text-anchored ink renders overlapping the typed text block; remarkable_read returns no typed text from the same v6 page
Summary
Two related problems on pages that mix typed text with handwriting (native notebooks, current firmware / v6, cloud mode):
- Rendered images draw the ink and the typed text block on top of each other. On the device, ink written below a typed block is anchored to a text character (
Group.anchor_id) and moves with it as the text reflows._v6_paths_from_blocksdraws strokes at their raw stored coordinates and never consults the group's anchor, so anchored handwriting lands ins
Read the thread · 2026-07-20 · closed · external user · 1 comment
Request dependency bump to resolve v5 stroke rendering and empty text layer errors via uvx
When running the server via uvx, Claude encounters an unhandled exception during the image rendering stage on modern reMarkable notebooks using the .v5 stroke format. The background stroke parser fails to compute the layout, resulting in empty text layers and an entirely frozen file state.Could the underlying rmc or rmscene library dependencies be updated in the package setup to explicitly support v5 rendering blocks out-of-the-box?
Read the thread · 2026-06-19 · closed · external user · 2 comments
The remaining reports are on the project's issue tracker.