OfficeCLI
OfficeCLI is the world's first and the best Office suite designed for AI agents.
Give any AI agent full control over Word, Excel, and PowerPoint — in one line of code.
Open-source. Single binary. No Office installation. No dependencies. Works everywhere.
OfficeCLI's built-in HTML rendering engine reproduces documents with high fidelity — and that's what gives AI eyes. It renders .docx / .xlsx / .pptx to HTML or PNG, closing the render → look → fix loop.
English | 中文 | 日本語 | 한국어
🌐 Website: officecli.ai | 💬 Community: Discord
PPT creation process using OfficeCLI on AionUi
PowerPoint Presentations
—
Word Documents
—
Excel Spreadsheets
All documents above were created entirely by AI agents using OfficeCLI — no templates, no manual editing.
For AI Agents — Get Started in One Line
Paste this into your AI agent's chat — it will read the skill file and install everything automatically:
That's it. The skill file teaches the agent how to install the binary and use all commands.
For Humans
Option A — GUI: Install AionUi — a desktop app that lets you create and edit Office documents through natural language, powered by OfficeCLI under the hood. Just describe what you want, and AionUi handles the rest.
Option B — CLI: Download the binary for your platform from GitHub Releases, then run:
This copies the binary to your PATH and installs the officecli skill into every AI coding agent it detects — Claude Code, Cursor, Windsurf, GitHub Copilot, and more. Your agent can immediately create, read, and edit Office documents on your behalf, no extra configuration needed.
For Developers — See It Live in 30 Seconds
That's it. Every add, set, or remove command you run will refresh the preview in real time. Keep experimenting — the browser is your live feedback loop.
Quick Start
Why OfficeCLI?
What used to take 50 lines of Python and 3 separate libraries:
Now takes one command:
What OfficeCLI can do:
Create documents from scratch -- blank or with content
Read text, structure, styles, formulas -- in plain text or structured JSON
Analyze formatting issues, style inconsistencies, and structural problems
Modify any element -- text, fonts, colors, layout, formulas, charts, images
Reorganize content -- add, remove, move, copy elements across documents
Word — full i18n & RTL support (per-script font slots, per-script BCP-47 lang tags lang.latin/ea/cs, complex-script bold/italic/size, direction=rtl cascading through paragraph/run/section/table/style/header/footer/docDefaults, rtlGutter + pgBorders shorthand, locale-aware page numbering for Hindi/Arabic/Thai/CJK; create --locale ar-SA auto-enables RTL), paragraphs (framePr, tabs shorthand, char-based indents), runs (underline.color, position half-pts), tables (virtual column ops add/remove/move/copyfrom, hMerge), styles, textbox / shape (textbox: rotation, textDirection eaVert/vert270, gradient, shadow, opacity), headers/footers, images (PNG/JPG/GIF/SVG), equations (LaTeX input), diagrams (mermaid → native editable shapes, or any mermaid type as a full-fidelity PNG), comments, footnotes, watermarks, bookmarks, TOC, charts, hyperlinks, sections, form fields, content controls (SDT), fields (22 zero-param types + MERGEFIELD / REF / PAGEREF / SEQ / STYLEREF / DOCPROPERTY / IF), OLE objects, revisions / tracked changes (revision.type=ins\|del\|format\|moveFrom\|moveTo + revision.action=accept\|reject, per-target /revision[@author=Alice] selector, tracked Find&Replace), page background color, document properties
Excel — cells (phonetic guide / furigana on add, Excel-UI --shift left\|up on remove / shift=right\|down on add), formulas (350+ built-in functions with auto-evaluation, spilling dynamic arrays with _xlfn. auto-prefix, financial / bond and statistical families, OFFSET/INDIRECT, defined-name formula bodies inlined at parse, formula-ref rewrite on row/col insert), sheets (visible/hidden/veryHidden, print margins, printTitleRows/Cols, RTL sheetView, cascade-aware sheet rename, empty-cell bloat filter on open), boolean and/or selectors (row[Salary>5000 and Region=EMEA]), tables, sort (sheet / range, multi-key, sidecar-aware), conditional formatting, charts (including box-whisker, pareto with auto-sort + cumulative-%, log axis), pivot tables (multi-field, date grouping, showDataAs, sort, grandTotals, subtotals, compact/outline/tabular layout, repeat item labels, blank rows, calculated fields, persistent labelFilter / topN filters, cache CoW + cross-pivot sharing), slicers, named ranges, data validation, images (PNG/JPG/GIF/SVG with dual-representation fallback), sparklines, comments (RTL), autofilter, shapes, OLE objects, CSV/TSV import, $Sheet:A1 cell addressing
PowerPoint — slides (header/footer/date/slidenum toggles, hidden), shapes (pattern fill, blur effect, hyperlink tooltip + slide-jump links, highlight color on runs, slideMaster/slideLayout typed add/set/remove, arrow alias, effective.X + effective.X.src), images (PNG/JPG/GIF/SVG, fill modes: stretch/contain/cover/tile, brightness/contrast/glow/shadow, rotation, link + tooltip), tables (built-in PowerPoint style catalogue, virtual /col[C] get + swap/copyFrom, row/col Move/CopyFrom, fill/background alias), charts (pieOfPie, barOfPie, per-attr axisLine/gridline setters, series add/remove with theme palette, anchor=x,y,w,h shorthand), animations (15 emphasis + 16 exit template-backed presets, multi-effect chains, motion-path presets, repeat/restart/autoReverse, chart animations + chartBuild), transitions (morph + p14 + 12 p15 PowerPoint 2013+ presets), 3D models (.glb) (combined rotation=ax,ay,az), slide zoom, equations (LaTeX input), diagrams (mermaid flowchart / sequence → native editable shapes, or any mermaid type as a full-fidelity PNG), themes, connectors (from/to accept a full /slide[N]/shape[@name=Foo] path), video/audio (loop, autoStart), groups (link + tooltip; Get/Query/Add/Remove all descend into groups), notes (RTL, lang), comments (RTL, legacy + modern p188 threaded round-trip), SmartArt (round-trip via add-part + raw-set), OLE objects, placeholders (add/set by phType)
Use Cases
For Developers:
Automate report generation from databases or APIs
Batch-process documents (bulk find/replace, style updates)
Build document pipelines in CI/CD environments (generate docs from test results)
Headless Office automation in Docker/containerized environments
For AI Agents:
Generate presentations from user prompts (see examples above)
Extract structured data from documents to JSON
Validate and check document quality before delivery
For Teams:
Clone document templates and populate with data
Automated document validation in CI/CD pipelines
Installation
Ships as a single self-contained binary. The .NET runtime is embedded -- nothing to install, no runtime to manage.
One-line install:
Or via a package manager:
Or download manually from GitHub Releases:
Verify installation: officecli --version
Or self-install from a downloaded binary (or run bare officecli to auto-install):
Updates are checked automatically in the background. Disable with officecli config autoUpdate false or skip per-invocation with OFFICECLI_SKIP_UPDATE=1. Configuration lives under ~/.officecli/config.json.
Key Features
Built-in Engines & Generation Primitives
OfficeCLI is self-contained. The capabilities below ship inside the binary — no Office required.
Rendering engine — high-fidelity, built-in
OfficeCLI's keystone: a from-scratch, high-fidelity HTML rendering engine that lets an AI agent see the rendered document instead of guessing from the DOM. It covers shapes, charts (trendlines, error bars, waterfall, candlestick, sparklines), equations (OMML → MathJax-compatible), 3D .glb models via Three.js, morph transitions, slide zoom, and shape effects. Per-page PNG screenshots are produced by piping the rendered HTML through a headless browser. Three modes:
view html — standalone HTML file, assets inlined. Open in any browser.
view screenshot — per-page PNG, ready for multimodal agents to read.
watch — local HTTP server with auto-refreshing preview; every add / set / remove updates the browser instantly. Excel watch supports inline cell editing and drag-to-reposition charts.
Without visualization, an agent generating slides is flying blind — it can read the DOM but can't tell if the title overflows or two shapes overlap. Because rendering is built into the binary, the render → look → fix loop works in CI, in Docker, on a server with no display — anywhere the binary runs.
Formula & pivot engine
350+ built-in Excel functions evaluated automatically on write — write =SUM(A1:A2), get the cell, the value is already there. No round-trip through Office to recalc. Covers spilling dynamic arrays (FILTER / SORT / UNIQUE / SEQUENCE / LET / LAMBDA / MAP), VLOOKUP / XLOOKUP / INDEX / MATCH, financial & bond math (XIRR / PRICE / YIELD / DURATION / COUPNUM), statistical distributions, tests & regression (NORM.DIST / T.TEST / LINEST), and date & text functions.
Plus native OOXML pivot tables from a source range with one command — multi-field rows/cols/filters, 10 aggregations, showDataAs modes, date grouping, calculated fields, top-N, layouts. Pivot cache + definition are written to OOXML, so Excel opens the file with the aggregation already populated:
Template merge — generate once, fill many
merge replaces {{key}} placeholders in any .docx / .xlsx / .pptx with JSON data — across paragraphs, table cells, shapes, headers, footers, and chart titles. Agent designs the layout once (expensive); production code fills it N times (cheap, deterministic, zero token cost). Avoids the failure mode where an agent regenerates each report from scratch and produces N inconsistent layouts.
Round-trip dump — learn from existing docs
dump serializes any .docx, .pptx, or .xlsx — whole document or any subtree (a single paragraph, table, slide, worksheet, the styles part, numbering, theme, or settings) — into a replayable batch JSON; batch replays it. Given a sample the user wants to imitate, an agent reads the structured spec instead of raw OOXML XML, mutates, and replays. Bridges "I have an existing template" and "generate me 100 variations."
Resident Mode & Batch
For multi-step workflows, resident mode keeps the document in memory. Batch mode applies multiple operations in a single pass.
Reading the file with another tool? Flush to disk first.
officecli's own reads (get/query/view) always see your latest edits, so within officecli you never need to save. But a live resident defers the disk write, so before a non‑officecli program reads the file — python‑docx/openpyxl, Microsoft Word, a renderer, delivery/upload — flush it:
A live resident also auto‑flushes shortly after going idle (adaptive 2–10s, scaled to the document's measured save cost). For a pipeline where another program reads after every command, set OFFICECLI_RESIDENT_FLUSH=each — every mutation is on disk before the command returns, while the resident stays warm. Full flush model (each/auto/fixed/off, save / close, env tuning): wiki → open / close.
Three-Layer Architecture
Start simple, go deep only when needed.
AI Integration
MCP Server
Built-in MCP server — register with one command:
Exposes all document operations as tools over JSON-RPC — no shell access needed.
Direct CLI Integration
Get OfficeCLI working with your AI agent in two steps:
Install the binary -- one command (see Installation)
Done. OfficeCLI automatically detects your AI tools (Claude Code, GitHub Copilot, Codex) by checking known config directories and installs its skill file. Your agent can immediately create, read, and modify any Office document.
If auto-install doesn't cover your setup, you can install the skill file manually:
Feed SKILL.md to your agent directly:
Install as a local skill for Claude Code:
Other agents: Include the contents of SKILL.md in your agent's system prompt or tool description.
Why your agent will thrive on OfficeCLI
Deterministic JSON output — every command supports --json with consistent schemas. No regex parsing, no scraping stdout.
Path-based addressing — every element has a stable path (/slide[1]/shape[2]). Agents navigate documents without understanding XML namespaces. (OfficeCLI syntax: 1-based indexing, element local names — not XPath.)
Progressive complexity (L1 → L2 → L3) — agents start with read-only views, escalate to DOM ops, fall back to raw XML only when needed. Minimizes token usage.
Self-healing workflow — validate, view issues, and the structured error codes (not_found, invalid_value, unsupported_property) return suggestions and valid ranges. Agents self-correct without human intervention.
Built-in agent-friendly rendering engine — view html / view screenshot / watch emit HTML and PNG natively. No Office required. Agents can see their output and fix layout issues, even inside CI / Docker / headless environments.
Built-in formula & pivot engine — 350+ Excel functions auto-evaluated on write (incl. spilling dynamic arrays, financial / bond and statistical families); native OOXML pivot tables from a source range with one command. Agents read computed values and shipped aggregations immediately, without round-tripping through Office.
Template merge — agent designs the layout once, downstream code fills {{key}} placeholders N times. Avoids burning tokens regenerating every report from scratch.
Round-trip dump — dump turns any .docx, .pptx, or .xlsx into replayable batch JSON. Agents learn from human-authored samples by reading a structured spec, not raw OOXML XML.
Built-in help — when unsure about property names or value formats, the agent runs officecli <format> set <element> instead of guessing.
Auto-install — OfficeCLI detects your AI tooling (Claude Code, Cursor, VS Code, …) and configures itself. No manual skill-file setup.
Built-in Help
Don't guess property names — drill into the help:
Run officecli --help for the full overview.
JSON Output Schemas
All commands support --json. The general response shapes:
Single element (get --json):
List of elements (query --json):
Errors return a non-zero exit code with a structured error object including error code, suggestion, and valid values when available:
Error codes: not_found, invalid_value, unsupported_property, invalid_path, unsupported_type, missing_property, file_not_found, file_locked, invalid_selector. Property names are auto-corrected -- misspelling a property returns a suggestion with the closest match.
Error Recovery -- Agents self-correct by inspecting available elements:
Mutation confirmations (set, add, remove, move, create with --json):
See officecli --help for full details on exit codes and error formats.
Comparison
Command Reference
End-to-End Workflow Example
A typical self-healing agent workflow: create a presentation, populate it, verify, and fix issues -- all without human intervention.
Units & Colors
All dimension and color properties accept flexible input formats:
Common Patterns
From Python or Node.js — install one of the thin resident-pipe SDKs (no per-call process spawn):
Both SDKs auto-provision the native CLI when missing (mirror-first, Windows-capable) and announce the install rather than doing it silently.
Or wrap subprocess directly, one-shot:
Documentation
The Wiki has detailed guides for every command, element type, and property:
By format: Word | Excel | PowerPoint
Workflows: End-to-end examples -- Word reports, Excel dashboards, PowerPoint decks, batch modifications, resident mode
Runnable examples: examples/ -- copy-paste scripts (.sh/.py) for Word, Excel, and PowerPoint, with output files included
Troubleshooting: Common errors and solutions
AI agent guide: Decision tree for navigating the wiki
Decision tree for navigating the wiki
Build from Source
Requires .NET 10 SDK for compilation only. The output is a self-contained, native binary -- .NET is embedded in the binary and is not needed at runtime.
License
Apache License 2.0
Bug reports and contributions are welcome on GitHub Issues.
If you find OfficeCLI useful, please give it a star on GitHub — it helps others discover the project.
OfficeCLI.AI | GitHub