A Microsoft Excel add-in (beta) from Anthropic that lets you ask Claude questions about a workbook, get cell-level citations, and make spreadsheet edits while preserving formulas.

What it is

Quick take: Claude in Excel is an Anthropic-built Excel add-in that turns “spreadsheet archaeology” into a conversation—with receipts (cell-level citations). It’s still beta, but it’s already the cleanest way I’ve found to interrogate a messy workbook without accidentally detonating a formula chain.

Gabriel’s notes

Claude in Excel is designed to understand an entire workbook (including multi-tab dependencies) and help you explain, debug, and update models while preserving formula integrity. Key features include: explanations with cell-level citations, scenario testing that highlights changes, and fast tracing of classic Excel gremlins like #REF!, #VALUE!, and circular references. It currently supports .xlsx and .xlsm files, and there’s a keyboard shortcut to pop it open quickly (Ctrl/Alt style, depending on OS).

I saved this under AI because it’s a practical example of LLMs doing real “knowledge work” in the exact place businesses already keep their truth (and lies): spreadsheets.

Context from my own workflow: I used Claude in Excel alongside Exa Websets (https://websets.exa.ai/) to clean up and enrich a tools/resources database before an initial upload—over 800 items. That’s not “AI replacing humans.” That’s “AI replacing the part of me that copies and pastes until my soul leaves my body.”

Good fit if you want to:

Understand what a workbook is doing (and why) without manually spelunking through 14 tabs.
Get formula explanations that point to specific cells, so you can verify instead of just believing the robot.
Run quick scenario tests by changing assumptions while keeping dependent formulas intact.
Debug spreadsheet errors fast and keep the fix localized (instead of “fixing” it by deleting the whole sheet).
Draft or populate financial models/templates when you already know the structure you want.

Pricing snapshot (auto-enriched):

Claude in Excel is included with paid Claude plans; Anthropic lists it as available (in beta) for Claude Pro, Max, Team, and Enterprise. Claude Pro is listed at $17/month when billed annually (or $20/month billed monthly). Enterprise pricing is Unknown / not confirmed (typically sales-led).

Work-use / compliance snapshot (auto-enriched):

Anthropic states Claude in Excel “works within your existing security/compliance framework,” but also flags the obvious: Claude can make mistakes, so you should review changes before finalizing—especially client-facing deliverables.

Important nuance: Anthropic’s consumer products (including Claude Free/Pro/Max) have a training/retention choice in privacy settings; if you allow data use to improve Claude, data may be retained for up to 5 years (per Anthropic’s Privacy Center). For Anthropic’s commercial products (e.g., Claude for Work / API), Anthropic says inputs/outputs are not used for training by default. If you’re doing client work, regulated work, or anything “please don’t leak this,” you should confirm which plan you’re on and what your org’s policy is before you feed it sensitive spreadsheets.

Alternatives (auto-enriched):

Microsoft 365 Copilot in Excel (Agent Mode): If your company is already standardized on Microsoft 365 Copilot, Agent Mode can do similar side-by-side workbook building/editing, and Microsoft notes Claude models may be selectable in Agent Mode (depending on tenant settings and licensing). It’s more “enterprise suite” than “single-purpose add-in.”
Gemini in Google Sheets: If you live in Google Workspace instead of Excel, Gemini in Sheets gives comparable “AI in spreadsheets” capabilities (cleaning data, generating formulas, creating tables/charts), with Google positioning it under Workspace’s enterprise security controls.

Before you adopt it:

Decide your data boundary. If it’s client-confidential, confirm plan type + privacy/training settings before you install and sign in.
Start with a sacrificial copy. Use a duplicate workbook for early experiments—especially with complex models and macros.
Force it to cite cells. Treat citations like unit tests: if it can’t point, it can’t prove.

Sources:

https://claude.com/claude-in-excel
https://claude.com/pricing
https://support.claude.com/en/articles/12650343-using-claude-in-excel
https://privacy.anthropic.com/en/articles/10301952-updates-to-our-privacy-policy
https://privacy.anthropic.com/en/articles/7996868-i-want-to-opt-out-of-my-prompts-and-results-being-used-for-training-models

Visit the resource