The spreadsheet plugin
for your agent.
Claude Code, Codex, Cursor or VS Code gets a spreadsheet it can calculate, check and fix. On your machine, beside you in the grid.
Open the workbook in VS Code. The agent highlights the range it means and asks you in the grid, not in a sidebar.
More accurate, on a public benchmark.
Agents are not bad at spreadsheets. They are bad at spreadsheets they cannot calculate. With SheetXL every write is recalculated and read back, so the agent verifies a result instead of hoping for one.
Same models, different harness. That is where the gap comes from.
How the run was madeSpreadsheetBench · verified-400
tasks solved · October 2026- SheetXL97%
- Grid Agent91.25%
- Shortcut.ai86%
- Claude Opus 4.6, bash agent28.75%
- Claude in Excel15.36%
- GPT-5.2, bash agent14.29%
SheetXL: our SpreadsheetBench submission, November 2026. The highest score of any vendor that has published its results. Other figures from the public leaderboard (V1 Verified), which also carries higher entries with no published outputs or method.
Lower costs. Fewer mistakes.
Fewer tokens let the agent focus on your problem, not the data.
- SheetXL3.3k
- Agent alone≈23.1k
Also fewer tool calls and faster per task, for the same reason.
Agent alone: Claude Code reading the file with Python, no engine. SheetXL measured per task over the same 400 SpreadsheetBench tasks, October 2026. The agent-alone figure is derived from Grid Agent's published savings over a bare Python agent and SheetXL's measured 3.0× over Grid Agent.
Works with the agent you already run.
Bring your Claude, Codex, Cursor or Copilot subscription. Your files stay on your machine.
Your AI
Claude Code, Codex, Cursor, Copilot. No credits to buy. You pay them for tokens, and nobody else.
Your machine
Your files never leave it. No second data path, no second vendor to put through security review.
Your expertise
The grid your people know, the agent your firm already approved. Productive on day one.
You already pay for your AI. Some AI spreadsheets bill you for it again. We don't.
Deterministic. Local. Reversible.
An agent you can let near the numbers, because the engine is in the loop and the workbook never leaves the machine.
Workbooks never leave the machine
The engine runs in your agent's process or a local server. No upload, no SaaS in the loop, no telemetry on your workbooks.
Every change is reversible
Sandbox sessions, undo, commits and an action log. The agent tries a fix, verifies it, and rolls back cleanly if it was wrong.
Excel-grade, verified
545+ functions checked against Excel. Dates, percents and errors are typed values, so what the agent reads is what it can write back.
Open and inspectable
MIT-licensed UI and skills. The harness that scored the benchmark is the one you install, not a private build.
What is in the box.
One plugin, four parts. Install any of them alone; together they are the whole product.
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. It highlights the cells it means and asks you in the sheet.
One engine, three layers.
A component for your app, a library for your server, a plugin for the agent you already run. Most spreadsheet vendors cover one layer.
| In your appA spreadsheet UI you embed | On your serverA calculation library for Node | With your agentA plugin for the agent you already run | |
|---|---|---|---|
| SheetXL | |||
| Univer | |||
| GRID | |||
| SpreadJS | — | — | |
| SheetJS Pro | — | — | |
| HyperFormula | — | — | |
| Coherent Spark | — | Their server | — |
| Handsontable | — | — | — |
| AG Grid | — | — | — |
| Shortcut | — | — | Their agent |
| Tracelight | — | — | Their agent |
Checked November 2026 against each vendor's public product and pricing pages. A check means a supported product for that layer that runs where you run. "Their server" and "Their agent" mean the vendor does it inside their own product: on their servers, or with their AI, not yours.
Per person, per month.
People pay per person. Trying it is free, and your files are never hostage.
Try it at work or at home
- Your AI, no token upcharge
- Limited agent runs
- Sends usage stats to SheetXL
Unlimited individual use
- Unlimited agent runs
- Sends usage stats to SheetXL
- Your AI, no token upcharge
For teams
- Unlimited agent runs
- Telemetry off, or routed to you
- Central licenses
Enterprise, for many people or your own telemetry endpoint: see pricing.
Your agents. Your spreadsheets. Your terms.
Pick your client. Every tab installs the same plugin; the first one is a prompt.
Paste 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 --allThen give me a quick intro to the new things you can do.