Sheetxl

SheetXL — the agent-first spreadsheet engine

The Excel-compatible spreadsheet engine your agents, your apps and your servers can run.

▶ Live demo · 📘 Docs · ⚡ Benchmarks · 🔌 API reference · 💲 Pricing · 💬 Discord

Table of contents

• See it running
• Just the facts
• One engine, three ways in
• For agents — a plugin
• For apps — a React component
• For servers — a headless library and CLI
• Compared
• The engine
• License and pricing
• Why SheetXL?
• Roadmap
• Credits
• Learn more

▶ Try the live demo — the full spreadsheet in your browser, no signup

Agents can reach every system except your spreadsheets. They read the file with Python, guess at what Excel would have calculated, and hope. SheetXL gives them the real thing: a calculation engine with 545+ Excel functions that an agent can query, change, recalculate and verify, with an audit log and undo, on your machine. The same engine is a React component for your app and a headless library for Node.

# No install. Ask a workbook a question from the terminal.

npx sheetxl -e "wb.getRange('Pricing!B14').getValue()" model.xlsx

Or paste this into Claude Code, Codex, Cursor or any agent that can run a shell:

I'd like to upgrade your spreadsheet skills with SheetXL:

npx -y skills add sheetxl/skills -g --all

Then give me a quick intro to the new things you can do.

*The agent highlights the cells it means and asks you in the grid, not in a sidebar.*

See it running

Each picture opens that workbook in the live demo. Nothing to install, and every cell is editable.

Sales dashboard. 600 orders behind KPI tiles, tables, a heatmap, charts , a pivot table , dynamic arrays and conditional formats .

Mandelbrot. One formula spills 8,192 cells and a color scale paints them. Change the zoom and it redraws.

Start from scratch. The whole app, a blank workbook and the template gallery. Paste from Excel, drop in an xlsx, or type a formula.

Storybook. Every component on its own, with the props to try. The quickest way to see what you would build with.

The workbooks are templates from the demo's New Workbook gallery; they live in sheetxl/templates-workbook. More in the demo gallery.

Just the facts

97% on SpreadsheetBench

Verified-400. Agents reading the file with Python score 14–29%.

7× fewer output tokens per task

The engine does the arithmetic, so the model does not.

< 0.4 s for a million formulas

Filled and calculated. Saved in about a second, in 160 MB.

545+ Excel functions

Dynamic arrays, LAMBDA, financials, statistics, text and dates.

120,000+ test fixtures

Each one checked against what Excel actually returns.

475+ shortcut keys and gestures

As in Excel, so your fingers already know them.

Got questions about a number? Ask us on Discord.

SpreadsheetBench: our submission, October 2026; baseline figures are from the public leaderboard (V1 Verified). Tokens: measured per task over the same 400 tasks; the agent-alone figure is derived, and the method is on sheetxl.com/agent. The harness that scored the benchmark is the one you install, not a private build. Engine timings and method are in the performance benchmarks.

One engine, three ways in

1. For agents — a plugin

Claude Code, Codex, Cursor or VS Code gets a spreadsheet it can calculate, check and fix. It is a spreadsheet plugged into your agent, so the agent that already has your code and your data gets the workbook as one more thing it can run.

• Your AI. Bring the subscription you already have; SheetXL never marks up tokens.
• Your machine. Your workbooks never leave it. No upload, no SaaS in the loop.
• Verified, not hoped. Every write is recalculated by a real engine and read back.
• Reversible. Sandbox sessions, a commit per change, undo, and an action log of what the agent did and why.

# Claude Code — the skill and the MCP server, at one version

claude plugin marketplace add sheetxl/skills claude plugin install sheetxl@sheetxl

Codex

codex plugin marketplace add sheetxl/skills codex plugin add sheetxl@sheetxl

VS Code — the grid, the engine and the MCP server. No Node, no terminal.

code --install-extension sheetxl.sheetxl-vscode --pre-release

Any MCP client

npx -y @sheetxl/mcp

What is in the box:

• The skill — teaches the agent to query a workbook like a database and run TypeScript inside the engine. MIT.
• The MCP server — sessions, commits, undo and an audit log. Finds the workbooks you have open.
• The CLI — the same engine from a shell. What the agent runs when there is no server.
• The VS Code extension — a grid beside the agent, so a person stays in the loop.

Agent quickstart →

2. For apps — a React component

An Excel-compatible spreadsheet inside your product. Not "Excel-like": every shortcut key and drag gesture, dynamic arrays, formatted copy/paste, charts and pivots. Real DOM, not a canvas, so every cell is an element you can style with CSS, inspect in devtools and test like any other web UI.

# A new project, pre-configured

npm create sheetxl

Or add it to an existing React app...

React 19 and MUI 9 are peers.

npm install @sheetxl/studio-mui @mui/material @mui/styled-engine @mui/system @mui/utils @emotion/react @emotion/styled react react-dom

import { Studio } from '@sheetxl/studio-mui';

export default function App() { // Studio fills its parent container. return (

); }

Not using React? @sheetxl/studio-vanilla is the same app, prebundled behind a plain JavaScript API. React guide →

3. For servers — a headless library and CLI

Pricing models, reports and reconciliations, run by your services, your schedules and your agents. No screen, no Excel install, no LibreOffice, no network call in the loop. Node 22+.

npm install @sheetxl/sdk @sheetxl/io

import { Workbook, WorkbookIO } from '@sheetxl/sdk';

import { IOPlugin } from '@sheetxl/io';

// Register the file format handlers (xlsx, csv, …) once at startup await WorkbookIO.install(IOPlugin);

const workbook = await WorkbookIO.read('pricing-model.xlsx'); workbook.getRange('Inputs!B2').setValues([[1250]]); const quote = workbook.getRange('Quote!B10').getValues(); await WorkbookIO.write(workbook, 'quote.xlsx');