For many of us, LLMs have become an essential tool. The more we use them, the more we want to do with them. Instead of asking questions in isolation, we want these tools to be able to answer questions based on the data we have. Since spreadsheets run the world, it’s only natural that we want LLMs like Anthropic’s Claude to have access to our spreadsheet data and calculations seamlessly.

At GRID, we’ve developed an API that enables you to interact with spreadsheets over HTTP. Imagine integrating this API into a conversation with Claude: you could simply ask "What were the total sales for April?" or "If we manage to reduce churn to 2%, how will that impact 2025 revenue?", and Claude would fetch the answer directly from your spreadsheet. This is precisely what’s possible by connecting Claude to a Model Context Protocol (MCP) server, allowing interaction with your spreadsheets via GRID’s API.

In this tutorial, we’ll extend Claude’s desktop app so that it can write to cells, read cell values, and evaluate formulas in your spreadsheets. Here’s how to set it up step by step.

What you need

Before diving in, gather the essentials:

1. Ensure you have Claude for Desktop installed and updated

2. Install Git and Node.js. Git is your trusty tool for getting the MCP server’s code from GitHub, while Node.js runs the JavaScript server on your machine

3. Sign up for GRID, upload a spreadsheet, and keep your GRID API key handy

Setting up the local MCP server

With the prep done, it’s time to set up a local Node.js server that understands how to speak to GRID’s spreadsheet API. Open your terminal and navigate to whichever directory you use for development projects. You might keep everything nicely organised in a Development folder, you might throw it all on your Desktop. Whichever it is, the important bit is that you know where you’ve put it.

Once you’ve settled on a location, you need to get a copy of the MCP server code. Run:

git clone https://github.com/GRID-is/claude-mcp.git

Head into the new project folder:

cd claude-mcp

And install the project:

npm install

As well as installing all the project’s dependencies, this will also build the server. After it’s finished you’ll have a new sub-directory named dist. A file in there, dist/index.js, is the entrypoint that will let Claude connect to GRID.

Configuring Claude to use the MCP server

With your server ready to go, you need to tell Claude how to use it. Fire up Claude for Desktop and look for an option in its menus to access Developer settings — on MacOS it’s in the main Claude menu, under Settings and in the Developer tab. There, you should see a button labelled “Edit Config”. Click it, and you’ll either open an existing config file or create a fresh one.

Alternatively, you can access the config file directly. On MacOS you’ll find it at ~/Library/Application\ Support/Claude/claude_desktop_config.json, while on Windows you'll find it at %APPDATA%\Claude\claude_desktop_config.json.

In that file, you need to include JSON that tells Claude that when it wants to do spreadsheet things, it should use this Node.js script and this API key:

Make sure to change /ABSOLUTE/PATH/TO/claude-mcp to the real location of the Git repo, and YOUR_API_KEY to your GRID API key.

Save the config file and make sure to restart Claude for Desktop. Once you’ve done that, you’ll see a small hammer icon to the bottom-right of Claude’s chat interface, as shown in the screenshot below. The “3” next to the hammer tells you that Claude has three new tools to play with, all for reading and writing spreadsheets via GRID’s API.

Click the hammer icon and to see full details of the three tools. These represent the endpoints your MCP server provides, the most important of which is the query_workbook endpoint. This allows you to read and write from spreadsheet cells. If you don’t see any new tools, double-check that the path in your config is accurate and you haven’t inadvertently told Claude to look in your cat photos folder.

Chatting with a spreadsheet

Now for the good part: actually asking a question and watching Claude spring into action. Initiate a new chat in Claude and try a simple prompt along the lines of:

Please tell me the value in cell C7 from workbook id XXX.

(XXX is the GRID spreadsheet id, which you can find in your list of spreadsheets.)

You should see Claude mention it’s about to run the query, and then ask for permission to talk to the MCP server (“Allow tool from ‘grid’ (local)?”). Click “Allow for This Chat”, and then Claude with talk to your MCP server, the server will talk to GRID’s API, grab the data, and return it. Voilà: the value in cell C7 will appear in your chat.

For an encore, you can have your MCP perform spreadsheet calculations as well, applying values to input cells and returning calculated results. You might say:

Change the value in cell C7 to 42, and tell me what’s in cell C9 afterwards.

Your server will do precisely that, returning the calculated values a moment later.

Troubleshooting

Of course, it’s all fun and games until something goes awry. If you don’t see any new tools in Claude after you restart, first confirm you’re running the latest version of Claude, that the path to your server file (the absolute location of claude-mcp/dist/index.js) is correct, and that the config file you're editing is indeed the right one. If you encounter authorisation errors, consult Claude's log directory (for instance ~/Library/Logs/Claude on MacOS) to see if any complaints have been logged. You'll also want to check your GRID API key is valid.

What next?

By connecting Claude to GRID’s spreadsheet API through an MCP server, you’ve made it possible to query, update, and analyse workbook data simply by asking questions. But this is only scratching the surface. If you want to take things further, here are a few ideas:

1. Develop richer LLM instructions: Going further than simple questions, you can provide more context in chats. For example, if cell C7 is labelled “Yearly sales target”, you should be able to use that label in chats (“What’s the yearly sales target?” instead of “What’s in cell C7?”). When you add a spreadsheet to GRID’s API environment, instructions for LLMs — including labels for cells — are auto-generated. To find them, select your spreadsheet in the API environment and click GPT setup in the sidebar.

2. Create a project to bundle instructions and spreadsheets: a Claude project combines instructions, cell labels, and information about your workbook into a tidy package. You can then chat within the project without needing to specify workbook ids or cell references, making your interactions easier to write and understand.

3. Explore advanced formulas and analysis: GRID is a full spreadsheet engine, supporting over 600 formula functions. This gives you powers beyond simply reading and writing cell values — it means Claude can evaluate formulas for complex calculations.

Combine all these and you can start to phrase queries in your own words: “What’s the average monthly revenue over the last two years, and in which months did we achieve the highest and lowest sales?” Using your MCP server Claude can interpret references, evaluate formulas, do almost everything Excel can do — all within a chat interface.

Wrapping up

With a few shell commands and a little bit of tinkering, you’ve transformed Claude for Desktop into a powerful tool for managing spreadsheets. Using a local MCP server and GRID’s API, Claude can fetch data, update cells, and perform complex calculations — all from within a conversational interface. Instead of rummaging through various windows in Excel, you can now ask questions and receive instant answers. It’s a small change that might leave your colleagues green with envy.

Happy querying!