How to Use Claude Across Excel and PowerPoint with Shared Context and Skills

Anthropic’s March 11 update makes Claude for Excel and Claude for PowerPoint work as one continuous workspace, so you can analyze data in spreadsheets, keep that context alive, and apply it directly to slide decks in the same session. The official announcement covers the release, and the practical setup details live in Anthropic’s docs for Excel and LLM gateway deployment. This walkthrough shows how to install the add-ins, configure shared context, use Skills and Instructions, and deploy the add-ins through an enterprise gateway.

What changed, and who can use it

The new capability is shared conversation context across all open Excel and PowerPoint files in a single session. Claude can read spreadsheet cells, write formulas, merge datasets, edit slides, and carry the same conversation across both apps.

This release is available in beta for Mac and Windows users on paid plans. Anthropic also says Skills in Excel and PowerPoint are available on all paid plans.

For Excel specifically, Anthropic’s help docs say Claude Sonnet 4.6 is available, and support pages from the same week also mention MCP connectors for sources like S&P Global, LSEG, Daloopa, Pitchbook, Moody’s, and FactSet. If your team already uses external data systems, that connector support matters. It turns Excel into the handoff point between enterprise data and presentation output. For background on MCP, see What Is the Model Context Protocol (MCP)?.

Installation and setup

If you are using Claude directly, install the Office add-ins first. Anthropic instructs admins to deploy “Claude by Anthropic for Excel” from Microsoft AppSource through the Microsoft 365 Admin Center. The same general deployment pattern applies to PowerPoint.

For individual setup, you need:

Requirement	Details
Plan	Paid Claude plan
OS	Mac or Windows
Apps	Microsoft Excel and PowerPoint
Add-ins	Claude for Excel, Claude for PowerPoint
Access mode	Direct Claude account or enterprise LLM gateway

If your organization manages Office centrally, ask IT to push the add-ins through Microsoft 365 Admin Center. That gives you a more consistent rollout and avoids per-user installation drift.

Once the add-ins are installed, open both an Excel workbook and a PowerPoint deck at the same time. Shared context only matters when both apps are open in the same active workflow.

Your first cross-app workflow

A good first test is a simple finance flow:

1. Open a workbook with historical financials.
2. Open a PowerPoint deck with an executive summary.
3. Start the conversation in Excel.
4. Ask Claude to analyze the workbook.
5. Switch to PowerPoint and continue the same conversation without repeating context.

Example prompts:

In Excel

Review this workbook and identify the key drivers of revenue growth, gross margin changes, and any anomalies in working capital. Summarize the findings in a form suitable for a management presentation.

Then, without restating the analysis:

In PowerPoint

Use the analysis from the workbook to update slides 5 through 8. Keep each bullet to one line. Add a concise takeaway at the bottom of each slide.

That is the core release feature. You are not passing text manually. Claude keeps the session context across the open files.

How to structure prompts for shared context

Shared context works best when you keep the conversation state explicit. Treat Claude like a stateful assistant, not a stateless prompt box.

Use this pattern:

Step	Prompt style	Purpose
Analyze	“Review the workbook and identify…”	Build context
Confirm	“Use that analysis to…”	Reuse context
Constrain	“Keep bullets to one line…”	Apply formatting rules
Refine	“Revise slide 7 for investor tone…”	Target edits

The same principle shows up in broader agent design. Context quality drives output quality. If you want a deeper treatment, Context Engineering: The Most Important AI Skill in 2026 is a useful companion.

Keep prompts grounded in specific ranges, sheets, or slide numbers when precision matters. For example:

Compare assumptions on the Revenue Build and Margin Bridge sheets. Then update slide 6 with a three-bullet summary and slide 7 with the revised valuation range.

Using Skills inside Excel and PowerPoint

Anthropic added Skills to both add-ins. Skills are reusable packaged workflows, and any Skills already configured in Claude on web or desktop, whether personal or organization-wide, are available in the add-ins automatically.

Anthropic also ships preloaded starter skills for financial analysis workflows in both add-ins.

Starter Skills in Excel

Skill type	What it helps with
Model audit	Formula errors, balance-sheet integrity
Template building	LBO, DCF, 3-statement models
Comparable analysis	Comps setup and review
Data cleanup	Messy spreadsheet normalization

Starter Skills in PowerPoint

Skill type	What it helps with
Competitive landscape decks	Build decks from source material
Deck updating	Refresh slides with new information
Banking deck review	Number consistency, narrative alignment, language polish

If your team is already using custom Skills, they should appear in the Office add-ins with no extra setup. Anthropic’s platform release notes describe Skills as packaged folders of instructions, scripts, and resources, and note that they can include executable code. If you are deciding how to package reusable workflows, What Are Agent Skills and Why They Matter and How to Create Your First Agent Skill are the relevant references.

A practical example:

In Excel

Run the comparable company analysis skill on this sheet. Use EV/Revenue, EV/EBITDA, and P/E where available. Flag missing or inconsistent peer metrics.

In PowerPoint

Use the comps analysis from this session to refresh the valuation summary slide and the peer positioning slide. Keep the existing deck style.

Instructions vs Skills

Anthropic separates Instructions from Skills, and that distinction matters in production workflows.

Use Instructions for persistent preferences. Use Skills for reusable task logic.

Examples of Instructions:

• Always use the firm’s number formatting in Excel
• Keep PowerPoint bullets to one line
• Flag cells referencing hardcoded assumptions

Examples of Skills:

• Build a DCF template
• Audit a three-statement model
• Review a banking deck for numerical consistency

This is similar to the distinction between global behavior rules and reusable procedures in developer tools. If you work with coding agents, Agent Skills vs Cursor Rules: When to Use Each maps closely to the same design choice.

A clean setup is:

Layer	Use for	Example
Instructions	Always-on preferences	One-line bullets
Skills	Repeated workflows	DCF build, comps review
Session prompts	Task-specific requests	Update slides 4-9 for Q2 data

Enterprise deployment with an LLM gateway

Anthropic also added support for running the add-ins through an internal LLM gateway backed by Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry. This is the same gateway pattern Anthropic uses for Claude Code, and Anthropic says teams already using that pattern can point the Office add-ins to the same endpoint with no new infrastructure.

End users need two values from IT:

• Gateway URL
• API token

Configuration happens inside the add-in settings pane. Open the Claude add-in in Excel or PowerPoint, navigate to the gateway settings, and enter the URL and token provided by your IT team. There is no client-side config file. The routing decision happens behind the gateway layer.

Gateway provider options

Provider	Good fit for	Notes
Amazon Bedrock	AWS-centric orgs	Aligns with existing Bedrock governance
Google Vertex AI	GCP-centric orgs	Useful if model routing already lives in Vertex
Microsoft Foundry	Microsoft-heavy enterprises	Fits existing Azure and Foundry controls

If your company already standardizes model access through a gateway, use that path. It gives you centralized auth, logging, governance, and model routing. If you are early in your agent infrastructure journey, AI Agent Frameworks Compared: LangChain vs CrewAI vs LlamaIndex is a useful overview of orchestration patterns around this layer.

Gateway mode limitations

Gateway mode does not support the full feature set available through a direct Claude account. According to Anthropic’s documentation, the following are not available when using an LLM gateway: Connectors (MCP), shared context across apps, Skills, file uploads, Memory, and web search (except on Vertex AI). If your workflows depend on Skills or cross-app context, you need direct Claude access rather than a gateway deployment.

Practical limits and tradeoffs

This release improves workflow continuity, but it still has constraints.

First, shared context is available in beta. You should expect occasional UX rough edges, especially in longer sessions with multiple open files.

Second, availability is limited to paid plans and Mac/Windows. There is no indication in the release that this is broadly available outside those environments yet.

Third, enterprise routing adds compliance and governance complexity. If you are using Microsoft’s Copilot-side Anthropic integration instead of Anthropic’s own add-ins, Microsoft documents a caveat that Anthropic-backed experiences are excluded from the EU Data Boundary and some in-country processing commitments. The March 2026 date in Microsoft’s documentation refers to the full availability rollout of Anthropic models within Microsoft 365, not to a resolution of the EU Data Boundary exclusion. Anthropic models remain disabled by default for EU/EFTA/UK customers. Security teams should review this before rollout.

Fourth, context continuity is useful, but it can also carry stale assumptions. In long sessions, reset or restate critical facts before making final slide edits. Context windows are bigger than they used to be, especially with Claude’s broader platform improvements, but stale or overloaded state is still an operational concern. Context Windows Explained: Why Your AI Forgets covers the mechanics.

A practical rollout plan for teams

For a small team, start with one narrow workflow that already jumps between Excel and PowerPoint, such as weekly KPI review, board deck refreshes, or comps updates.

Define:

1. One workbook template
2. One deck template
3. A short set of persistent Instructions
4. One or two approved Skills
5. A review checkpoint before final export

That setup gives you repeatability without overbuilding. It also matches the way Anthropic framed the release, as a workflow upgrade for spreadsheet-to-deck work rather than a standalone model feature.

If your team uses external financial data, test Excel with one MCP-backed source first, then layer PowerPoint output after the spreadsheet workflow is stable. If your team uses enterprise routing, wire the add-ins to your existing Bedrock, Vertex AI, or Foundry gateway before expanding to more users.

Start with one analyst workflow that already moves from workbook to deck every week. Install both add-ins, set 2 to 3 persistent Instructions, try one of the preloaded financial analysis starter skills, and measure whether your team produces the same deck with fewer manual handoffs over the next seven days.