Which server to choose

Two paths depending on where the data is.

  • For spreadsheets (Excel, .xlsx): a dedicated Excel server. It reads and writes cells, formulas, new sheets, and in some cases creates charts and pivot tables. It's the right choice if you work with sheets.
  • For structured data in a local database (a SQLite file, for example): a database server. More suitable when the data is plentiful and organized into proper tables.

If your "database" is actually a giant Excel sheet, you need the Excel server, not the database one.

How to do it

  1. Add the Excel server to the configuration file. The working syntax:
{
  "mcpServers": {
    "excel": {
      "command": "npx",
      "args": ["-y", "@negokaz/excel-mcp-server"]
    }
  }
}

This server reads and writes Excel files without needing the Excel program installed.

  1. Restart the assistant completely.
  2. Point to the file and ask. Tell the AI the path of the sheet and what you want: "open the file /Users/yourname/sales.xlsx, sum the revenue column by month and write the totals in a new sheet".
  3. Check the result. Reopen the sheet and verify the cells written. For formulas, check that the calculation is the one you wanted before trusting the totals.

A concrete example

Marco keeps his employees' hours in an Excel file and every end of month loses an afternoon doing the math. He connects the Excel server and points to the file. Then he asks: "read the June hours sheet, calculate the total for each person, flag who exceeded 160 hours and write a summary in a new sheet called Summary".

The AI opens the file, does the calculations, writes the summary sheet with the totals and overtime highlighted. Marco reopens the file and spot-checks two totals: they add up. The afternoon of calculations has become a minute of asking and two of checking. The file never left his computer.

When it does NOT work (and how to fix it)

If the file is open in Excel

If the sheet is open in the program, the server may fail to write to it (the file is locked). Close Excel before asking the AI to modify it, then reopen it to see the result. It's the most common cause of a "can't save".

If complex formulas don't add up

On intricate calculations the AI can get a formula wrong. Ask it to "write the formula in the cell, not just the result": that way the sheet calculates on its own and you see and check the formula. For the calculations that matter, always verify a couple of cells by hand.

If the file is huge

Sheets with tens of thousands of rows can slow down or not be read in full. Ask the AI to work on a precise range ("rows 1 to 500 of the Orders sheet") or work on a reduced copy with only the data you need right now.

A tip from someone who actually uses it

Always work on a copy of the sheet, at least until you trust it. An AI writing into an Excel file can overwrite a cell by mistake, and on a real working file that would be trouble. Keep the original safe, have the AI operate on the copy, and transfer the results only once you've verified they're right.

Frequently asked questions

Do I need to have Excel installed on the computer?

No: these servers read and write .xlsx files on their own, even if you don't have the Excel program. You only need Node.js to run the server. You then open the resulting sheet with any program that reads Excel files, even a free one.

Can the AI create charts and pivot tables?

Some Excel servers allow it (charts, pivot tables, formatting), others limit themselves to reading and writing values. Check what the server you choose offers on its page; if you need a specific feature, verify it's among those supported before counting on it.

Does my Excel data end up online?

The file stays and is processed on your computer by the local server. What you ask it to analyze, however, passes through the AI assistant: the sheet isn't uploaded anywhere, but the data that goes into your questions and answers travels through the model. For sheets with very sensitive data, keep that in mind.