mrkeyoor.com_
Tue 29 Sept 21:20 UTC
Webevaluationupdated 28 Sept 2026

m3e-canvas review

M3E Canvas is a browser editor for sketching Material 3 Expressive phone and desktop screens, connecting their interactions, and turning the result into a prompt for an AI coding tool. It solves the awkward gap between describing an interface in prose and building a full design file before the first code draft.

trackingstars / 7d
Verdict

Our M3E Canvas run installed 352 packages, built successfully, and passed all 750 tests, so it is an easy recommendation for solo developers who want a Material 3 screen brief before coding. Use it for early interaction sketches and AI handoff, especially when local storage and static hosting are benefits. Choose a full design platform when the deliverable must support team editing, editable vector handoff, or production-ready code.

We ran it

Lab card: what happened when we ran m3e-canvasScreenshot of m3e-canvas (lnkiai.github.io/m3e-canvas)
Install✓ · 19s352 packages · 563 MB
Build✓ · 16s
Tests✓ · 9s750 passed · 0 failed of 750 (vitest)
Known vulns00 critical · 0 high · 0 moderate · 0 low (npm audit)
Repo108 files~30,766 lines of source · 10 MB · 2 CI workflows

Answers from our run

Does m3e-canvas build from source?

Dependencies installed in 19 seconds (352 packages), and the build succeeded in 16 seconds. We cloned commit de2d97a into a clean Debian container with 3 CPUs and no project-specific setup.

Do m3e-canvas's tests pass?

Yes: 750 of 750 passed when we ran the project's own test command (vitest). Some failures need services or credentials a bare container does not have.

Does m3e-canvas have known vulnerabilities in its dependencies?

npm audit found none in the dependency tree at the time of our run.

Who should not use m3e-canvas?

Product teams that need live multi-user editing, accounts, comments, or server-side version history: the project has no backend, and documents live in browser local storage or files.

What are the alternatives to m3e-canvas?

Penpot, Excalidraw, tldraw. Our M3E Canvas run installed 352 packages, built successfully, and passed all 750 tests, so it is an easy recommendation for solo developers who want a Material 3 screen brief before coding.

Setup5/519-second install; build and all 750 tests passed
Docs5/5Setup, security, contribution, and agent formats are documented
Community4/58,374 stars and four active issues or PRs after a recent push
Maturity3/5Clean tests, but no GitHub release and some handoff features are beta

Who it’s for

Android and web developers who want to settle screen structure before asking an AI coding tool to implement it.
Solo builders who prefer a local, account-free sketching tool with project files and share links.
Teams exploring Material 3 Expressive themes, motion, navigation, and responsive screen variants.
Contributors comfortable maintaining UI and prompt text in English, Japanese, Chinese, and Korean.

Who it’s NOT for

Product teams that need live multi-user editing, accounts, comments, or server-side version history: the project has no backend, and documents live in browser local storage or files.
Designers who require an editable Figma, Sketch, or Illustrator handoff: the README documents prompt and PNG output, while open issue 447 requests a structured SVG export.
People expecting the phone editor to match the desktop editor: the README limits mobile use to one fixed screen and button editing.
Organizations that cannot keep provider credentials in browser storage: the optional AI helper stores the user's API key in localStorage and sends requests directly to the chosen provider.
Anyone expecting generated production code: M3E Canvas exports a brief for a coding tool, and the repository does not claim that the resulting app is verified.

Setup reality

Our sandbox install succeeded in 19 seconds, adding 352 packages and using 563 MB on disk. The build passed in 16 seconds. Vitest then passed all 750 tests in 9 seconds, and npm audit reported 0 known vulnerabilities.

Local development needs npm and a supported Node release; package metadata starts at Node 22.12. The editor itself needs no account or backend. Only the optional AI helper needs a provider key, which stays in browser localStorage and goes directly to OpenAI, Anthropic, Google, or DeepSeek.

The production build is a static export. Hosting beneath a subpath requires NEXT_PUBLIC_BASE_PATH, while GitHub Pages is already configured for that case. The full multi-screen editor is for desktop browsers; the phone interface is a one-screen, buttons-only editor.

Two screen sizes become one implementation brief

M3E Canvas lets one design carry 412 x 892 phone screens and 1280 x 800 desktop screens, then describes matching names as two widths of the same screen. You drag Material 3 Expressive controls onto a canvas, connect taps or swipes, preview the transitions, and copy the result as an Android or web prompt. Theme controls cover color, shape, type, contrast, and motion. The purpose is narrow and useful: give a coding agent a concrete interface brief before it starts inventing layout.

That focus is visible in the 108-file repository we measured. It contains about 30,766 lines of source rather than a server, database, or plugin system. The component set covers bars, buttons, cards, lists, dialogs, fields, pickers, progress indicators, image placeholders, and other common Material parts. Prompts and interface text exist in English, Japanese, Chinese, and Korean. A contributor who adds a part has to update rendering, editing, prompt wording, and all 4 languages.

The handoff is a prompt or PNG, not an editable design file

M3E Canvas produces a prompt that you can edit before copying, plus a PNG of a screen. It also saves the whole project as JSON. Share links compress that document into the URL fragment, so the design is carried by the link and does not pass through an application server. Picked local images and AI rewrite history are removed from the shared copy, while HTTPS image references can remain. This is a tidy handoff for one builder or an async review.

The limit matters more than the feature count. Open issue 447 asks for SVG because the current PNG is a flat bitmap with no editable text, layers, or vector icons. Our 10 MB checkout grew to 563 MB after npm installed its dependencies, yet the output still stops before production code or a design-system artifact. If your next step is Figma refinement, token management, or a formal developer handoff, Penpot is a closer fit.

What happened when we ran it

Our sandbox installed 352 npm packages in 19 seconds and used 563 MB on disk. The Next.js build completed in 16 seconds. Vitest then finished in 9 seconds with 750 passed and 0 failed. Those numbers come from commit de2d97a in a fresh unprivileged container with 3 CPUs and 8 GB of RAM. For a visual editor with its own prompt and geometry logic, that is a clean first run.

Npm audit reported 0 known vulnerabilities in the installed tree. The repository had 2 CI workflow files, no Dockerfile, and no dedicated tests directory; its test files sit beside components and library code. Nothing in our run required a database, system package, or secret. Our test method covered installation, the production build, the available Vitest suite, and dependency audit. It did not score generated layouts or compare an AI-built app with the sketch.

No backend means no team workspace or secret vault

The security model is easy to understand because there are 0 application servers in the normal path. Drawings remain in localStorage, project files, or URL fragments. There are no accounts. That cuts deployment work and keeps a shared design out of a hosted project database, but it also means there is no shared workspace, access control, comment thread, or server-side revision history. Teams must supply those practices elsewhere.

The optional AI helper changes the privacy calculation. A user enters an OpenAI, Anthropic, Google, or DeepSeek key, and the browser stores it under m3e:ai. The generated description of the whole design goes directly to that provider. None of this affected the 16-second static build, and you can leave the helper off. A company that forbids client-side provider keys should do exactly that, or omit the feature from its deployed copy.

The full editor is desktop-only

On a phone, M3E Canvas gives you 1 fixed screen and a buttons-only editor. The multi-screen canvas, full part palette, grouping, layers, and broader editing flow belong on desktop. This is sensible for detailed layout work, though the README's "phone-friendly" label can sound broader than the actual mobile feature set. Mobile users can adjust a quick button sketch; they cannot replace a desktop design session.

The open queue shows where the edges still are. Issue 432 requests charts, issue 443 seeks feedback on the rebuilt customization panel, and issue 447 requests structured SVG export. GitHub listed 4 combined open issues and pull requests on September 28, 2026, while the last push landed on September 27. The project is moving quickly, and the small open queue includes specific product work instead of an abandoned backlog.

Active development has not produced a GitHub release yet

The repository was created on September 2, 2026 and had 8,374 stars by September 28. It has no GitHub release, while its package metadata says version 0.1.0 and marks the package private. Share links and agent-authored drafts are labeled beta. Those details do not make the editor unreliable, especially beside 750 passing tests, but they do make file compatibility and workflow stability worth checking before a team standardizes on it.

M3E Canvas earns a place before implementation, where a rough verbal request needs visible screens and explicit navigation. Our install finished in 19 seconds, the build in 16 seconds, and all 750 tests passed in 9 seconds, with 0 known vulnerabilities reported. That is strong evidence for trying it. The stopping point is equally clear: once several people need to edit the same source of truth, or the handoff must stay editable outside the browser, move the work into a full design platform.

Alternatives

ProjectWhat it isPick it when
Penpot gh↗A self-hostable interface design platform built around team collaboration and developer handoff.pick this instead when editable design files, shared libraries, comments, and team workflows matter more than prompt output.
Excalidraw gh↗A collaborative whiteboard for quick diagrams and hand-drawn interface sketches.pick this instead when you need a loose shared canvas for flows and discussion rather than Material 3 controls.
tldraw gh↗A React SDK and canvas application for building custom infinite-canvas experiences.pick this instead when the canvas itself is part of the product you are building and you need an SDK.

What people are saying

  1. [velocity-scout] lnkiai/m3e-canvas

Sources

  1. M3E Canvas repository and README
  2. M3E Canvas security model
  3. M3E Canvas contribution guide
  4. SVG export request
  5. Customization panel feedback issue
  6. Chart support request
  7. M3E Canvas releases page

More web reviews

vinext · robinhood-meme-token-stock-launchpad · kool-brushez · pure · shadcn-admin · fable51-worlds · the whole board →