Step-by-Step Guide: Connect Microsoft Excel to Claude in 2026

There are many ways to get Excel data in front of Claude, and most of them fall into two camps. The first is uploading a file, which works fine until the data changes and you’re uploading it again. The second is a live, secure, and governed connection to wherever your Excel files already live, so Claude reads the current workbook directly and the data is always up to date.

This guide covers the second path. It’s the one worth setting up if your team is doing real analysis rather than one-off questions, and it means nobody has to pass spreadsheet copies around to get an answer. It also doesn’t require anything exotic on the Claude side. CData Connect AI is a listed connector in Claude’s own directory, so there’s no MCP server URL to track down or paste anywhere.

Overview of Connecting Excel to Claude

Uploading a file into a Claude chat is the fastest way to ask one question about one spreadsheet. It also means Claude is looking at a frozen snapshot. The moment the source data changes, that snapshot is wrong, and someone has to remember to re-upload it.

A live connection (like the one you get through CData Connect AI) works differently. Claude queries the workbook directly, in place, every time you ask something. There’s no copy sitting in a chat thread somewhere, and no step where a person has to remember to refresh anything.

This matters most for workbooks that don’t stay still, such as a live pipeline tracker, a weekly inventory count, or a rolling expense sheet that finance updates every few days. A one-time budget template someone built last year is probably fine as an upload. Anything a team actually opens on a schedule is a better candidate for a connection than a chat attachment.

The rest of this guide walks through setting up live Excel connectivity, starting in the Connect AI, then inside Claude.

Step 1: Connect Excel for the web to CData Connect AI

Connecting to Excel Online in Connect AI takes just a few minutes.

1. In Connect AI, open the Sources page and click + Add Connection.

2. Search for “Excel Online” and select it.

3. On the Basic Settings tab, give the connection a name, or keep the default.

4. Choose an authentication method. For most individual or small-team setups, that’s AzureAD. Enter a SharePoint URL if you’re connecting to a SharePoint site, or leave it blank to connect to OneDrive instead.

5. Click Sign in, log in through the OAuth prompt, and grant the requested permissions.

6. Click Save & Test. A successful test sets the connection status to Authenticated, and you can view the resulting data model under the Data Model tab.

AzureAD covers most individual accounts and small-team setups. Larger organizations managing this at scale have two other options, AzureServicePrincipal and AzureServicePrincipalCert.

Both authenticate through a registered Azure app instead of an individual sign-in, using a tenant ID, client ID, and either a client secret or a certificate. The setup steps are close to identical. The main difference is that an admin registers the app once, and the connection doesn’t depend on any one person’s login staying active.

Before you start asking questions, note that in some configurations, workbook tables show up under the schema name OneDrive rather than ExcelOnline. If a query against a table you expect to see comes back empty, browse your schemas in Explorer, and it’s usually just a naming difference.

Step 2: Add the CData Connect AI Connector in Claude

With the Excel source connected on the Connect AI side, the next step happens entirely inside Claude, and it doesn’t involve pasting in an MCP server URL or editing any configuration file.

1. Sign in to claude.ai, open the left navigation menu, and go to Settings > Connectors.

2. Click the Discover tab in the top right to open the connectors directory.

3. Search for “CData Connect AI” and select it.

4. Click Connect to sign in and complete the authorization.

5. Click Configure to set permissions on individual tools, choosing between Always allow, Needs approval, or Block for each one.

The tool configuration step is the first real governance decision a user makes, and it’s a per-tool setting rather than an all-or-nothing switch. A read-heavy tool used for pulling data might stay on “Always allow,” while anything that writes or modifies data is a reasonable candidate for “Needs approval,” at least until you’ve seen how it behaves.

For a team just getting started, setting everything to “Needs approval” for the first week and loosening specific tools once you’ve watched Claude use them is a reasonable, low-risk default.

Step 3: Prepare Your Workbooks for AI Queries

Connect AI reaches workbooks stored in OneDrive or on a SharePoint site, so keep the files you want Claude to query there. Clean structure also improves the quality of Claude’s answers. Here’s a quick data readiness checklist.

1. One header row with consistent column names.

2. Standard date and number formats.

3. No merged cells.

4. No blank rows breaking up a data range.

None of that requires restructuring a workbook from scratch, just tidying the habits that tend to creep in over months of manual edits.

A workbook with three different date formats in the same column, or a summary row inserted halfway down a data range, is the kind of thing that trips up a formula too. Claude isn’t any more forgiving of it than Excel is.

Step 4: Query Your Excel Data with Claude

Once the connector is configured, start a new chat and ask a question. Claude explains which Connect AI tools it’s using to answer, and if a tool is set to Needs approval, it asks before running it.

From there, the questions can be as specific as you’d ask a colleague. Here are a few that work well against a typical operating spreadsheet.

• “Show me month-over-month variance on this expense tab, and flag anything that moved more than 15%.”

• “Summarize the top 10 accounts by revenue in this workbook.”

• “Which rows in this inventory sheet look like outliers compared to the rest of the column?”

CData’s prompt library has more examples if you want a starting point for your own team’s use cases.

Step 5: Save and Share Claude’s Results

Claude can turn what it finds into new, formatted spreadsheets or reports that you can download and share. Saving results as new files rather than overwriting the source keeps a clean trail of what changed and when. It also leaves the original workbook intact for anyone else who relies on it.

Claude can also write back to the original workbook. Through Connect AI, the Excel Online connector supports inserting, updating, and deleting rows, as long as the connection’s permissions allow write operations. Each change applies to the workbook immediately, with no undo step. Test write-back on a copy of the workbook first, and keep write permissions limited to the users and connections that need them.

Advanced: Combine Excel with Other Enterprise Data

The same connection can do something a single Excel file can’t: combine the workbook with other connected systems in one prompt. If your organization connects Salesforce, SAP, or another source through Connect AI, Claude can compare data across systems without anyone exporting a second file first. For example:

• Budget against pipeline: Compare a budget workbook against live Salesforce pipeline data.

• Inventory against orders: Check an inventory sheet against current order volume.

• Budget against actuals: A finance team can reconcile a quarterly budget tracker against NetSuite actuals by asking for the comparison directly. There’s no need to pull a NetSuite export and line the two up by hand.

It’s the same governed connection on both ends, just pointed at two systems instead of one.

This setup also works beyond Claude. The same Connect AI connection works with other AI assistants, including ChatGPT and Microsoft Copilot. A team that standardizes on Connect AI for governance keeps its choice of AI vendor open.

Best Practices for Secure and Compliant AI Integration

The governance story here comes from two layers working together.

The first is OAuth passthrough. When Claude queries a workbook through Connect AI, it’s using the signed-in user’s own OneDrive or SharePoint credentials, not a shared service account with broad access. If someone doesn’t have permission to open a file directly, Claude querying on their behalf doesn’t change that. The access is theirs, and it’s enforced exactly where it already was.

The second layer is the tool-level permissions set up in Claude’s connector settings. Even within the access a user already has, Always allow, Needs approval, and Block let a team decide which actions should run automatically and which should pause for a human to confirm. Query logging on the Connect AI side means there’s a record of what was asked and retrieved, which matters when a security team needs to answer “what did the AI actually touch” rather than guess.

Put together, the two layers answer two different questions. OAuth passthrough answers “could this person see this data at all.” Tool permissions and logging answer “what did Claude actually do with the access it had.” Neither one on its own is a complete governance story, but together they cover most of what a security review is going to ask about.

Here’s a quick checklist for rolling this out responsibly.

1. Connect only the OneDrive folders or SharePoint sites a given team actually needs, not an entire tenant.

2. Set any write-capable tools to Needs approval or Block until you’ve watched how they behave.

3. Review query logs periodically rather than only after something goes wrong.

Connect your Excel data the governed way with CData Connect AI

Uploading a file to Claude answers one question about one moment in time. A live connection through CData Connect AI answers the question that comes after it, and the one after that, against data that’s still current and permissions that never left the source system.

A 14-day trial is available if you want to set this up against your own OneDrive or SharePoint data directly.

Frequently Asked Questions