MCP Server for Google Sheets: the options

The right MCP server for Google Sheets comes down to one decision: service-account auth or your own OAuth login. If you want the widest Sheets tool surface — read, write, format, batch, charts — install mcp-google-sheets and point it at a Google service account. If you'd rather log in as yourself and also touch Drive files, use mcp-gdrive with an OAuth client. That's the whole shortlist for Google Sheets; the rest of this page is which to pick, the config that works, and the three "Google" servers people install by mistake.

Both of these run locally over stdio, which puts them in the roughly 90% of MCP servers that are a local process on your own machine rather than a hosted endpoint. For a Google credential, that's the setup you want — the token never leaves your laptop.

Pick by how you want to authenticate

Start from auth, not features. A service account is a robot identity you create in Google Cloud and share a specific spreadsheet or folder with; OAuth logs the server in as you. That single choice picks your server.

Server	Best for	Auth	Transport	Official?
mcp-google-sheets	Full Sheets tooling on shared/team sheets	Service account (or OAuth)	stdio	Community
mcp-gdrive	Sheets cells + Drive search, as yourself	OAuth (client ID/secret)	stdio	Community

One-line rule: automating team spreadsheets or a shared folder → mcp-google-sheets with a service account. Working across your own Drive and reading/writing the odd sheet → mcp-gdrive. Neither is an official Google product, so treat both as community servers and scope their access tightly.

mcp-google-sheets: the default for real spreadsheet work

For anything beyond reading a few cells, mcp-google-sheets is the server to reach for. It's a Python server that exposes read, write, formatting, batch-update, and chart tools — the closest thing to "drive Sheets from an agent" available today. It runs over stdio via uvx, so there's nothing to build.

The cleanest setup uses a service account. Create one in Google Cloud, download its JSON key, enable the Sheets and Drive APIs, then share the target spreadsheet (or a folder) with the service account's email. The server only sees what you share — that sharing boundary is your access control.

{
  "command": "uvx",
  "args": ["mcp-google-sheets@latest"],
  "env": {
    "SERVICE_ACCOUNT_PATH": "/full/path/to/service-account-key.json",
    "DRIVE_FOLDER_ID": "your_shared_folder_id_here"
  }
}

The key file is a real credential, so keep it out of any committed config and store the path via your client's env handling. Scope it by sharing only the folder the agent needs — not your whole Drive. The server also supports plain OAuth if you'd rather act as yourself, but for team or automation use, the service account is the less-surprising choice.

mcp-gdrive: when Sheets is part of a Drive workflow

Reach for mcp-gdrive when Google Sheets is one stop in a broader Drive task — search for a file, read a doc, then read or write some sheet cells. It lists, reads, and searches Drive files and can read and write Google Sheets cells, all as you, via an OAuth client. It's a TypeScript server that runs over stdio with npx, so again nothing to compile.

The trade-off versus mcp-google-sheets: you get Drive breadth but a thinner Sheets surface. mcp-gdrive handles cell read/write; it isn't the tool for formatting rules or building charts. The setup cost is also higher — you create an OAuth client in Google Cloud and pass its ID and secret, plus a directory where the server caches your token.

{
  "command": "npx",
  "args": ["-y", "@isaacphi/mcp-gdrive"],
  "env": {
    "CLIENT_ID": "<CLIENT_ID>",
    "CLIENT_SECRET": "<CLIENT_SECRET>",
    "GDRIVE_CREDS_DIR": "/path/to/config/directory"
  }
}

First run opens a browser to authorize; the token lands in GDRIVE_CREDS_DIR for reuse. Since the OAuth grant is scoped to your account, review the scopes it asks for and treat that credentials directory like an SSH key.

The three "Google" servers to skip for Sheets

Most people connecting Sheets pick the wrong server because it has "Google" in the name. Here's what each actually does, so you don't install it hoping for spreadsheet tools:

• MCP Toolbox for Databases (BigQuery) is Google's official server, but it's for BigQuery, not Sheets. If your data lives in BigQuery datasets, this is the right tool; for a spreadsheet, it does nothing.
• Google Calendar MCP (nspady) manages Calendar events across accounts. Same Google login story, entirely different surface — no cells, no sheets.
• Chrome DevTools MCP drives a real Chrome and can, in theory, click through the Sheets web UI. Don't. Automating the browser to edit a spreadsheet is slow, brittle, and burns tokens on DOM noise when an API-backed server does the job cleanly.

The pattern: match the server to the API, not the brand. "Google" in a name tells you the vendor, not the capability.

What to install and how to wire it up

Install one Sheets server, not both. They overlap on cell read/write, and every extra server eats into the roughly 40-tool budget most clients handle before tool selection gets unreliable — the Cursor tool limit math spells out why that ceiling is real. Pick mcp-google-sheets for full spreadsheet control, or mcp-gdrive if you mostly live in Drive.

Adding either is the same three steps in any client — command, credentials, restart — walked through in how to add an MCP server. To compare the wider field or filter by what a server can do, start from the best MCP servers, browse capabilities, or check alternatives if neither Sheets server fits your workflow.