explainx.ai350K+ learn from us
TrendingAI News TodayExplainx for Business
Pricing
explainx.ai

Upskill in AI — 16 free pathways, live workshops & bootcamps, and 50+ courses from practitioners. Plus the skills, tools, and MCP servers to practice on.

follow us

follow on google

Add explainx.ai as a preferred source

corporate training

support@explainx.ai

get started

Find your pathTake Free Evaluation

community

Join the community

learn

mind: share how you thinkpathways — start freeworkshopsbootcampscoursescompare Explainxcertificationsmock testsexplainx universitycorporate traininglearn skills & mcp

discover

skillsmcp serversexplainx mcpestacktoolsmdx readeragentsllmsdesignsdictionarypeopleagi trackerfelony benchranks

company

aboutvisionmissionteaminstructorsteach on explainxpartnershipscommunityhackathonscareers

content

daily AI newsstate of AI — live resultsblogreleasespromptsgeneratorsresource libraryfor LLMsexplainx.ai kids

solutions

all solutionsdeveloper upskillingmarketing upskillingproduct manager upskillingleadership upskilling

newsletter · weekly

Get AI news, tools, and insights in your inbox.

supportcontactprivacytermsdata rightshow we create contentsubmission guidelines

© 2026 AISOLO Technologies Pvt Ltd

explainx.ai

On this page

  • TL;DR
  • What is in the repository?
  • How to run it locally
  • MCP for AI agents
  • Embed vs AGPL self-host
  • npm packages
  • OAuth and developer console
  • Architecture: what is open vs closed
  • Package map for monorepo readers
  • Spreadsheet semantics for agents
  • Presentation and PDF surfaces
  • Desktop vs browser-only
  • Star velocity and maintenance risk
  • Comparison table: MROffice vs alternatives
  • What people are asking on Hacker News
  • Security and data handling
  • Limitations (honest)
  • Bottom line
  • Related reading
← Back to blog

explainx / blog

MROffice: Open-Source Office Editors, MCP Tools, and What AGPL Means for Agents

Open Source, AI Agents, MCP, Tools

Part of AI Tools and Apps

MROffice is an AGPL office-editor monorepo with docs, sheets, slides, PDF, ink, and code packages, plus mroffice-mcp for AI apps. How to try it, embed vs self-host, and license risks.

Nov 22, 2026·9 min read·Yash Thakker
add explainx.ai
go deep
MROffice: Open-Source Office Editors, MCP Tools, and What AGPL Means for Agents

MROffice is a fast-moving open-source office editor stack — about 1,800 GitHub stars within roughly twelve days of its November 10, 2026 creation date per the repo header — that targets builders who want Microsoft Office-like editing in the browser without shipping a closed SDK. The project separates AGPL editor source from a hosted embed and desktop shell that reads and writes .docx, .xlsx, and .pptx.

This post explains what shipped in the repo, how agents hook in via MCP, and where the license bites.

TL;DR

table · 2 cols
QuestionAnswer
LicenseAGPL-3.0 plus NOTICE.md attribution to mroffice.ai
EditorsDocs, sheets, slides, PDF, ink, code — packages under @mrnote/*
Agent pathmroffice-mcp or hosted https://mcp.mroffice.ai/mcp
Embed pathnpx mroffice-embed init + server createSessionToken
Native Office filesHosted embed or desktop, not pure OSS build
Try without clonemroffice.ai live demo
Daily briefFrom Yash · 350K+ students taught

Catch up on AI

One AI story explained in plain English, plus one thing to try. Every day, free. Unsubscribe anytime.

What is in the repository?

Open and closed document editor stacks linked by a cable, illustrating MROffice embed versus AGPL self-hostOpen and closed document editor stacks linked by a cable, illustrating MROffice embed versus AGPL self-host

Per the README:

Six editor packages

table · 3 cols
EditorOpen formatsNotes
Documents.mrdoc, HTML, Markdown, textStyles, tables, tracked changes UI
Spreadsheets.mrsheet, CSV, TSVCell math in browser; function formulas kept as text in OSS
Presentations.mrslidesThemes, transitions, animations preview
PDFRead, annotate, printSave returns unmodified bytes in OSS
Handwriting.mrinkPressure-sensitive ink
CodeMany languagesRuns JS/TS, Python, SQL, Lua, shell in browser

Charts, SmartArt layout, and function evaluation (SUM, VLOOKUP, etc.) run in the proprietary MR Office engine inside the desktop app, per the README. OSS builds preserve those objects but do not recalc everything locally.

Apps and tooling

Three fanned document sheets, illustrating the six-editor office stackThree fanned document sheets, illustrating the six-editor office stack

  • apps/playground — browser demo at localhost:5178 after npm run build && npm run preview.
  • apps/desktop — Electron shell; optional MR Office component for full Office binaries.
  • apps/mcp — source for mroffice-mcp.

How to run it locally

From the README's quick path:

bash
npm install
npm run fonts
npm run build
npm run preview

npm run dev serves the same playground for development. npm run smoke checks a build.

Fonts: npm run fonts installs open substitutes for Office fonts — required before a faithful preview.

MCP for AI agents

MCP connectors linking tools to an agent workflowMCP connectors linking tools to an agent workflow

The README documents two agent entry points:

  1. Hosted MCP — https://mcp.mroffice.ai/mcp (sign in with an MR Office account, no API key in the quick start).
  2. Local MCP — npx -y mroffice-mcp on a developer machine.

That pattern matches how explainx.ai covers other document agent bridges — see Univer's facade API and AWS agent skills for caching. MROffice's pitch is real editor sessions, not headless conversion only.

When to pick MCP vs npm packages

table · 2 cols
ApproachBest for
mroffice-mcpAgent frameworks that already speak MCP (Claude Desktop, custom hosts)
mroffice-docs npmReact apps embedding a single editor component
mroffice-embedSaaS products that need .docx without AGPL on your core app

Embed vs AGPL self-host

Hosted embed (recommended in README)

Your server mints a short-lived session token with createSessionToken from mroffice-embed/server. The browser calls open() from mroffice-embed. Files round-trip through signed callbacks.

Free tier limits quoted in README (November 2026): 1,000 sessions/month on localhost, 100 on custom domains, up to 2 apps.

Because MR Office hosts the editor, your product is not automatically AGPL when you use the hosted service — the README explicitly contrasts this with shipping modified AGPL sources yourself.

Self-host AGPL editors

If you modify and serve the six editors over a network, AGPL-3.0 requires offering complete corresponding source to users, plus NOTICE.md branding rules (keep MR Office attribution and link).

A commercial license from support@mr-note.com is the escape hatch if you need proprietary distribution.

npm packages

Install per editor, for example:

bash
npm i mroffice-docs mroffice-core

Sibling packages: mroffice-sheets, mroffice-slides, mroffice-pdf, mroffice-ink, mroffice-code. Each README documents React embedding.

compare to Gamma agent slides when your agents only need deck generation, not Excel-grade grids.

OAuth and developer console

MR Office also acts as an OAuth 2.1 authorization server (PKCE, dynamic registration). That is separate from MCP but useful if your product wants Sign in with MR Office instead of storing Office credentials.

Developer docs live at mroffice.ai developers per the README (sign-in required).

Architecture: what is open vs closed

The README's diagram is explicit:

  • AGPL block: @mrnote/core, six editors, playground, desktop shell source, and MCP server source.
  • Non-OSS component: MR Office component that implements Word/Excel/PowerPoint/PDF write-back and advanced engine features — free to use but not open source.

That split is why the project can be trending on GitHub without giving away the Office compatibility layer. It mirrors how OnlyOffice-style products separate community document editors from document server builds.

Session flow for embed

  1. User authenticates on your app.
  2. Your backend calls createSessionToken with appId, secretKey, file metadata, and user id.
  3. Frontend open() loads the hosted editor with the token.
  4. On save, MR Office POSTs to your signed callback with the updated file.

This is closer to Figma embed or Google Picker than to shipping a static iframe of a bundled WASM build. Agents using MCP skip your iframe but still rely on MR Office's hosted runtime unless you run mroffice-mcp locally with a desktop install.

Package map for monorepo readers

table · 2 cols
PathRole
packages/docsDocument editor React component
packages/sheetsSpreadsheet grid
packages/slidesDeck editor
packages/pdfViewer and markup
packages/inkHandwriting canvas
packages/codeLightweight IDE surface
packages/coreShared UI and model code
fonts/office-packOpen font substitutes
llms.txtMachine-readable project summary for agents

The repo ships llms.txt, signaling that maintainers expect coding agents to read the tree — similar spirit to AGENTS.md in other trending repos.

Spreadsheet semantics for agents

OSS spreadsheets evaluate arithmetic between cells (=A1*B1+10) but keep function formulas as written without full Excel recalc. Agents that overwrite cells can therefore leave stale function results until the proprietary engine runs.

Mitigation patterns:

  • Use hosted embed when agents must return customer-grade Excel.
  • For internal dashboards, export CSV after agent passes and re-import through a pipeline you control.
  • Log before/after cell JSON when agents touch sheets — auditability matters for finance agent posts.

Presentation and PDF surfaces

Slides support eight themes, master/layout views, and animation preview in OSS. PDF mode supports highlight, underline, strikethrough, squiggly, and comments but does not write annotations back into the PDF bytes on OSS save paths.

For teams comparing Gamma-generated decks vs live slide edits, use Gamma 5 coverage for generation, MROffice for pixel-level deck surgery.

Desktop vs browser-only

npm run desktop launches MR Office Editors shell. Without the proprietary component, desktop mode still opens .mrdoc / .mrsheet / .mrslides and PDFs. With the component from Releases, it becomes the full MR Office desktop app with AI features and native Office binaries.

Windows and macOS installers are advertised at mroffice.ai for teams that need offline access — relevant when your agent must edit on an air-gapped machine and sync later.

Star velocity and maintenance risk

~1,755 stars in roughly twelve days (per candidate feed) puts MROffice in the same attention bucket as other agent-tooling repos explainx.ai covered this month. High stars do not guarantee:

  • Long-term maintenance
  • Stable MCP schemas
  • Pen-tested embed callbacks

Pin commit SHAs in production integrations and run npm run smoke in CI after upgrades.

Comparison table: MROffice vs alternatives

table · 4 cols
ProductLicense / hostNative .docxAgent hook
MROffice OSSAGPL self-hostNonpm + optional MCP
MROffice embedHosted commercialYesSame MCP + tokens
UniverApache-2.0 facadePartial via integrationsFacade API
Google Docs APISaaSYesREST, not MCP-native
Microsoft 365 CopilotSaaSYesMicrosoft graph stack

Pick two layers: one generation tool (LLM) and one faithful editor (MROffice embed or desktop) when contracts require Office file types.

What people are asking on Hacker News

The repo's HN thread (low score, early stage) will matter less than agent builders trying embed limits. Common questions map to:

Is this AGPL contamination for my SaaS?

Use hosted embed if you do not want copyleft on your core codebase. Self-host only with legal review.

Can agents replace LibreOffice headless?

Different tradeoff. MROffice optimizes WYSIWYG and human+agent co-editing. Headless LibreOffice suits batch conversion; see Libreboot anti-AI policy discourse for another angle on OSS purity vs agents.

How is this different from Univer?

Univer focuses on facade APIs for spreadsheets and docs your agents already control. MROffice adds MCP, PDF ink, and a commercial Office-IO layer. Many teams will prototype on Univer and pilot MROffice embed when .docx fidelity matters.

Security and data handling

  • Session tokens must stay server-side; README shows mro_sk_ secrets only on the server.
  • Callbacks are signed — treat forged save URLs like any webhook auth bug.
  • AGPL does not imply security audit; run your own review before customer PII.

For agent misuse patterns, see OpenRouter support-email read claims as a reminder that tools that read documents need scope limits.

Limitations (honest)

  • Star velocity ≠ production maturity — the repo is only weeks old; expect API churn.
  • No native .docx in OSS — demo videos may show desktop/hosted features; label your integration correctly.
  • Function recalc in sheets requires desktop/proprietary engine for full Excel parity.

Bottom line

MROffice is worth a look if you build agents that must edit real office artifacts and you can accept AGPL or paid embed. Clone for UI experiments; buy or embed for Word/Excel fidelity.

Related reading

  • Univer Office SDK for agents
  • What are MCP servers
  • ChatGPT accounting plugins
  • Agentes.chat self-hosted Slack alternative
  • Amazon Textract document AI GA
  • Gamma 5 agent presentations

Repository layout and license text are accurate as of November 22, 2026 per GitHub README.

Spotted something out of date? Let us know.
Daily briefFrom Yash · 350K+ students taught

Catch up on AI

One AI story explained in plain English, plus one thing to try. Every day, free. Unsubscribe anytime.

  • ✓ Free
  • ✓ One email a day, no spam
  • ✓ Unsubscribe anytime
Yash Thakker

Written by

Yash Thakker

Yash is an AI expert with over 300K learners. Join his workshops →

View Yash Thakker in People in AI →

Related posts

Oct 4, 2026

Top 10 MCP Servers in 2026: 6 Open Source, 4 Closed Source (and What to Check)

The MCP ecosystem is crowded, and the first question for any server is not what it can do but who controls it. This ranking splits ten useful MCP servers into six open-source and four closed, hosted ones, explains how we ranked them, and gives a checklist for connecting any of them safely.

Nov 20, 2026

agentes.chat: Self-Hosted Team Chat Where AI Agents Are Teammates

agentes.chat, trending on GitHub with about 1.8k stars, is a self-hosted team chat where AI agents have roles, owners and monthly budgets. Here is what it does, how to install it with Docker Compose, the AGPL terms, and what is still beta.

Nov 17, 2026

DIYgod Pinky: MCP Agents That Drive Your Phone With You in the Loop

With more than 2,200 GitHub stars in its first weeks, Pinky lets macOS, Windows, or Linux hosts run MCP-connected agents against a real phone—USB, Wi-Fi, or Tailscale—with checklists for OEM-specific Android debugging and on-device companion apps.