mrkeyoor.com_
Fri 04 Sept 06:44 UTC
Webevaluationupdated 04 Sept 2026

editor review

Pascal Editor is a browser-based 3D building editor and a set of embeddable TypeScript packages for scene editing, viewing, capture, and IFC conversion. Its local CLI also pairs the editor with an authenticated MCP server so coding agents can inspect and change the same projects.

Verdict

Our Pascal Editor run installed 1,049 packages and built in 151 seconds, but its viewer test task failed, so the codebase is capable and still carries integration risk. Try the one-command local editor if you want agent-controlled building scenes or a modern WebGPU interface. Pin versions and run your own browser, export, and MCP checks before embedding the beta packages in a product.

We ran it

Lab card: what happened when we ran editorScreenshot of editor (editor.pascal.app)
Install✓ · 108s1049 packages · 1288 MB
Build✓ · 151s
Tests✗ · 12sran, no count parsed
Repo2922 files~487,288 lines of source · 107.6 MB · 3 CI workflows · Dockerfile

Answers from our run

Does editor build from source?

Dependencies installed in 108 seconds (1049 packages), and the build succeeded in 151 seconds. We cloned commit 19327a9 into a clean Debian container with 3 CPUs and no project-specific setup.

Do editor's tests pass?

The test command failed in our container, and its output did not report a pass or fail count.

Who should not use editor?

Teams that need every monorepo test green before adoption: our run ended with @pascal-app/viewer#test failing after 9 of 16 tasks succeeded.

What are the alternatives to editor?

Sweet Home 3D, Bonsai and IfcOpenShell, Open BIM Components. Our Pascal Editor run installed 1,049 packages and built in 151 seconds, but its viewer test task failed, so the codebase is capable and still carries integration risk.

Setup3/5One-command local path; repository install is 1,288 MB and tests failed
Docs5/5Local CLI, packages, plugins, MCP, and storage are explained clearly
Community5/521,864 stars with August issue activity and a September push
Maturity3/5Useful packages exist, but beta integration and data bugs remain open

Who it’s for

Web teams building custom architecture, floor-plan, or building-data interfaces with React and Three.js.
Designers who want a local visual editor backed by SQLite instead of a mandatory hosted account.
MCP users who want Codex, Claude Code, or another compatible agent to edit structured scenes.
Plugin authors who need custom node types, renderers, placement tools, and editor panels.

Who it’s NOT for

Teams that need every monorepo test green before adoption: our run ended with @pascal-app/viewer#test failing after 9 of 16 tasks succeeded.
External apps importing the current beta editor package without integration work: open issue 715 reports that beta.5 pulls a Node-only module into optimized browser builds.
Local users who depend on the scan-upload button today: issue 645 documents that the self-hosted editor accepts a file and then does nothing.
Workflows that require lossless save/load of collections: issue 734 says the round trip silently drops them.
Strict MCP clients that reject JSON Schema draft-07 output declarations: issue 696 reports every tested tool call failing in Claude Desktop.

Setup reality

Our sandbox installed 1,049 packages in 108 seconds and used 1,288 MB. The build succeeded in 151 seconds. Tests failed after 12 seconds: the log named @pascal-app/viewer#test, with 9 successful tasks out of 16 and 6 cached.

The shortest local route needs Node.js 22.13 or newer, npm, and a browser. The CLI downloads a versioned runtime, starts the editor and authenticated MCP service on loopback ports, and stores projects in local SQLite. Repository development uses Bun and Turborepo across workspaces.

The CLI calls macOS its initial supported release; Ubuntu has automated smoke coverage, while broader Linux and Windows support is still being verified. External package consumers face a reported beta.5 browser-build break, and the local scan uploader has a documented no-op path.

The 2,922-file monorepo is an editor and a toolkit

Pascal Editor covers more than drawing walls in a browser. Its packages split scene schemas, WebGPU rendering, editing controls, built-in node types, capture sessions, IFC conversion, command-line management, and MCP access. The core scene is a flat node graph stored through Zustand, while render systems update dirty geometry in React Three Fiber. The standalone Next.js editor assembles those pieces for people who want the finished interface.

That scope has a real footprint. Our checkout held 2,922 files, about 487,288 source lines, and 107.6 MB before installation. Published use also requires several packages: the viewer guide installs core, viewer, editor, and nodes, plus peer dependencies such as Next.js, React, Three.js, React Three Fiber, Drei, and Zustand. The built-in plugin must load before the Viewer mounts or scene nodes have no definitions to render. Pascal makes extension possible, but the package boundary is something an integrator must understand.

Node.js 22.13 starts a local editor and MCP service

Node.js 22.13 or newer is the documented floor for the one-command local install. Running npx @pascal-app/cli editor downloads a versioned runtime, selects available loopback ports, checks editor and MCP health, and opens a .localhost URL. Projects live separately in ~/.pascal/data/pascal.db, so updating the runtime does not replace them.

Local security choices are sensible. Both services bind to 127.0.0.1, and MCP uses a random token kept in a private runtime directory rather than copied into client settings. pascal mcp setup codex and pascal mcp setup claude create stable connectors that discover the changing port. The MCP package can also run headlessly under Node.js 22.13 or Bun with local SQLite and no browser, WebGPU, React, or external database. Binding HTTP beyond loopback requires a bearer token and an allowed origin.

What happened when we ran it

Our sandbox installed Pascal Editor in 108 seconds, adding 1,049 packages and consuming 1,288 MB. The Turborepo build succeeded in 151 seconds. Tests then exited with code 1 after 12 seconds. This was a fresh unprivileged container with 3 CPUs, 8 GB of RAM, Node 22, and no secrets, using commit 19327a9. The repository included 3 CI workflow files, a Dockerfile, a compose file, and monorepo workspaces, but no top-level tests directory.

The failing log tail identifies @pascal-app/viewer#test and nothing more specific. It reports 9 successful tasks out of 16, with 6 cached, before Turbo exited. Other summaries show 16 capture-protocol tests, 9 IFC-converter tests, and 25 capture-viewer tests ran; a core hit-test case also passed. We will not guess which viewer assertion or setup step failed because the supplied tail does not say. The decision-grade finding is that this checkout built, then its full test command did not pass in our container.

Editor beta.5 can pull Node code into browser builds

Open issue 715 reports that @pascal-app/editor beta.5 imports a Manifold print-export path that eventually reaches node:module. The reporter's optimized Next.js and Webpack build then fails before browser runtime, even though ordinary editor checks work. That is narrow enough to act on: teams embedding the public editor entry point should reproduce their production bundle before upgrading.

The 1,288 MB installation also warns against treating Pascal as a drop-in widget. The viewer expects a WebGPU-capable browser and an explicit plugin load. The editor adds print export, direct manipulation, floor-plan state, and more package edges. Issue 617 records a consumer type-check failure tied to noUncheckedIndexedAccess, while issue 715 shows a server-only dependency reaching a browser build. Package users should pin a tested set of versions and build a small external host in CI rather than relying only on this monorepo's own build.

Local scan upload and save/load still have data gaps

Version 0.1.5 of the local CLI is named in issue 645, where choosing a GLB scan produces no request, progress indicator, or scene node because the local runtime does not supply the upload callback. Issue 733 describes local scan or guide files remaining in IndexedDB after their last scene reference is deleted. Issue 734 covers the opposite failure: collections disappear during Save Build and Load Build because neither export nor validation carries them through. These are specific edges, yet they touch the trustworthiness of local project data.

The failed 12-second suite raises the cost of dismissing those reports as isolated. A buyer should test the same operations used in production: save and reload a representative project, delete large local assets, export floor plans, import an IFC sample, and reconnect the chosen MCP client. Issue 629 supplies one IFC slab that imports with the wrong offset and thickness. Issue 696 reports strict JSON Schema validation rejecting local MCP tool output in Claude Desktop. Pascal covers many workflows, and each boundary deserves its own acceptance fixture.

September activity is ahead of the v0.9.1 release page

The repository had 21,864 stars and 38 open issues and pull requests when fetched. An issue-only search found 26 open issues, several updated on August 31, 2026. The last push was September 2. GitHub's latest-release endpoint returned v0.9.1 from June 10, while current issue reports refer to later beta packages. That combination shows continuing work after the named release, so the June date alone is poor evidence of neglect. It also makes version pinning more important because main, stable release notes, and beta packages describe different checkpoints.

Pascal Editor earns a serious trial for teams that want a programmable building editor in the browser, especially when MCP access to the same local SQLite scenes is useful. The 151-second successful build shows the monorepo can assemble in a plain container. The 1,049-package install, failed viewer test task, and open data-path bugs stop us from calling it a safe dependency upgrade without local checks. Use the CLI for a low-cost evaluation, then make the exact browser and file workflows you depend on pass before committing.

Alternatives

ProjectWhat it isPick it when
Sweet Home 3DA long-running desktop and web home-design application with its own project format.pick this instead when you want an established end-user home planner and do not need Pascal's TypeScript or MCP layer.
Bonsai and IfcOpenShellAn IFC toolkit whose Bonsai add-on provides native IFC authoring inside Blender.pick this instead when IFC is the system of record and a Blender-based professional workflow is acceptable.
Open BIM ComponentsThree.js components for building browser-based BIM viewers and tools.pick this instead when you want lower-level BIM components and plan to design the application shell yourself.

What people are saying

  1. [lobsters] Fine, I’ll build my own text editor
  2. [velocity-scout] pascalorg/editor
  3. [velocity-scout] decodethehassan/ELDEN-RING-NIGHTREIGN-Save-Editor-Edit-Money-Items-Stats-Free-
  4. [velocity-scout] soundernarayanasamy/Path-of-Exile-Character-Editor-Free-Download
  5. [velocity-scout] christiandlbanzon/The-Binding-of-Isaac-Rebirth-Save-File-Editor-Free-PC-Download
  6. [velocity-scout] zituhossain/Total-War-WARHAMMER-III-Character-Editor-Edit-Your-Character-Save-Free-

Sources

  1. Pascal Editor repository
  2. Pascal Editor v0.9.1 release
  3. Pascal CLI documentation
  4. Pascal viewer documentation
  5. External browser build issue
  6. Self-hosted scan upload issue
  7. Save and load collections issue
  8. MCP output schema issue

More web reviews

tdesktop · stylex · brave-browser · desktop · xterm.js · alpine · the whole board →