Open-source CLI that lets AI agents create and edit .docx, .xlsx and .pptx files
Your agent can write code all day but still cannot fill in the quarterly workbook; OfficeCLI is a single binary that closes that gap without a Microsoft Office install anywhere in the pipeline.
Narrow but genuinely useful scope, no cost, and a fast commit cadence with 23.8k stars, held back by a thin public track record and a hard limit to the three standard OOXML formats.
This is for teams whose agents already touch the filesystem and now need to produce real Office artifacts: reports, workbooks, decks, generated headlessly in CI or a container. It is free, Apache 2.0, actively committed to, and wired for Claude Code, Cursor, VS Code and Copilot, which makes it an easy yes if your deliverables end in .docx or .xlsx. Treat write access as the risky part, not the install.
Drop one executable on macOS, Linux or Windows and skip the usual Office automation stack entirely.
Built-in calculation means spreadsheet edits recompute instead of leaving dead cells behind.
Dump an existing document to a readable structure so an agent can learn its layout before writing changes.
Merge data into Word, Excel and PowerPoint templates with one consistent workflow.
High-fidelity HTML rendering gives you a way to preview or publish output without opening Office.
Documented use with Claude Code, Cursor, Visual Studio Code and GitHub Copilot rather than a bespoke SDK.
The binary can update itself, which matters given the project is shipping tags at a rapid clip.
Concrete setups pulled from the research, not feature-list hand-waving.
Data engineer on a reporting team
Add OfficeCLI to the CI container, pull the numbers from your warehouse or API, and merge them into an approved .docx template on a schedule.
The monthly report stops being someone's Friday afternoon and becomes a build artifact.
Agent tooling engineer
Expose OfficeCLI to Claude Code or Cursor on a scratch copy of a real workbook, first read-only via round-trip dump, then with write access on a narrow set of sheets.
The agent updates formulas and tables with recalculated values instead of handing back a CSV you have to paste in.
Solutions or sales engineer
Wire a prompt-to-pptx step that fills a branded template with the customer's specifics before a call.
A first-draft deck in the house format that only needs editing, not building.
Ops lead sitting on a shared drive
Point the binary at a directory of contracts or spreadsheets and extract structured fields into your own store.
Searchable structured data out of files nobody has opened in two years.
Open source, single binary, no dependencies beyond the .NET runtime it ships against
insightsFree and open source for capability that usually means either a licensed Office install or a hand-rolled openpyxl script, so the value question is your engineering time, not the invoice.
Pricing as captured on 2026-08-01. Check the live site before you commit.
Real reactions surfaced during research. Paraphrased faithfully, linked to source.
graphic_eqToo new for a real street reaction: the launch posts and writeups are warm, but the research surfaced no substantial independent user threads, no HN comment content, and no verified complaints.
A post on X reacts with straight astonishment and frames it as the first Office suite built for AI agents.
A LinkedIn post frames it as someone open sourcing an entire Office suite, which captures the launch-day tone more than any measured evaluation.
A hands-on writeup notes the scope is strictly .docx, .xlsx and .pptx, with no support for HWP or HWPX documents.
Lower-level Python libraries for reading and writing the three OOXML formats.
PICK IT WHENYou want maximum control inside an existing Python codebase and do not need a single-command, agent-facing interface.
Microsoft's own enterprise APIs for programmatic document and tenant access.
PICK IT WHENYour documents live in Microsoft 365 and you need tenant identity, permissions and audit trails rather than local offline edits.
Broader document parsing and transformation tooling across many file types.
PICK IT WHENYour job is ingesting and normalizing mixed documents, not writing back native Office files.
timerTime to first value: about 15 minutes
Download the single binary for macOS, Linux or Windows from the GitHub releases and confirm it runs.
Copy a real workbook or report to a scratch directory and use the round-trip dump to inspect its structure read-only.
Wire the binary into your agent of choice (Claude Code, Cursor, VS Code or Copilot) and grant write access only on the scratch copy first.
No. The whole point is a single binary that reads and writes the standard Office formats without an Office install, which is what makes it viable in Docker and on build servers. It does rely on the .NET runtime.
There is a built-in formula and pivot engine, so calculations are handled rather than left stale. Verify against a copy of your own workbook before trusting it on anything financial.
It is open source at $0 with no paid tier published, and writeups note the Apache 2.0 license. The cost is ownership: you integrate it, you test it, and there is no vendor support line.
The repository shows over 6,000 commits, 144 tags and commits landing within the last day, alongside 23.8k stars and 1.6k forks. Shipping pace looks strong; long-term stability is still unproven.
Install it, run a read-only pass against a copy of a workbook you actually care about, and only then let an agent write. The download costs nothing; a mangled financial model costs plenty.
Field research: 1 pages scraped · 3 search passes · 13 community sources. Reviewed by the Koda desk on 2026-08-01.