Case Study · AI Integration

Custom Claude Integration for Google Workspace — Google Sheets MCP Saves Hours Going Back and Forth

How to move from constant CSV downloads and copy-paste friction to Claude reading, writing, and creating Google Sheets autonomously

Pranoti Kshirsagar · AI Integration & Automation Specialist · thesciencetalk.com

5 steps

eliminated from every data task

0 €

ongoing cost

All sheets

accessible — not just one

01 · The real problem — death by data friction

The goal

Deep integration of Claude across all business workflows. Not a one-off connection — a flexible, reusable bridge between Claude and any Google Sheet, for any task: grants database, social media analysis, webinar tracking, client data.

The daily reality

Every time data needed to move between Claude and a spreadsheet, it required a full manual loop — downloading, converting, uploading, sharing, then copying results back out. Slow, error-prone, and capacity-draining.

02 · Before vs after — the workflow

Before

Export CSV from source

›

Download to local machine

›

Upload to Google Drive

›

Open as Google Sheet

›

Copy-paste into Claude

After

Tell Claude what to do with the sheet

›

Claude reads, writes, or creates directly

›

Done

02b · The hidden misconception

Common assumption that's wrong

Claude.ai has a native Google Drive connector — so surely it can access Google Sheets? No. The native Drive connector is limited to Google Docs only. Claude cannot read cell contents, write to cells, or create new spreadsheets through it. For any real Sheets integration, a dedicated MCP server is required — in Claude Desktop, configured separately from claude.ai entirely.

03 · Blockers and solutions

Blocker

Solution

Service account key creation blocked — Google Workspace org policy had JSON key downloads disabled. Every standard tutorial assumes this works.

✓

Switched to OAuth 2.0 Client ID (Desktop app type) — bypasses org policy entirely, uses personal auth instead of service account keys.

Google Drive API not enabled — mcp-google-sheets requires both Sheets API and Drive API. Enabling only one causes silent failures with no useful error message.

✓

Explicitly enabled both APIs in GCP Library — Sheets API and Drive API. They don't auto-enable each other.

uvx not resolving in Claude Desktop config — using "uvx" as the command caused silent launch failures with no error output to debug from.

✓

Replaced with the full absolute path to uvx (e.g. /Users/name/.local/bin/uvx). Resolved immediately.

04 · What Claude can now do with Google Sheets

✓

Read cell contents directly — no upload, no copy-paste

✓

Write and update cells without manually touching the sheet

✓

Create brand new spreadsheets from scratch on command

✓

Works across all sheets — grants database, social media analysis, webinar tracking, and more

05 · Final setup — what was built

Architecture

Google Cloud project — Sheets API + Drive API both enabled explicitly

OAuth 2.0 Client ID (Desktop app) — consent screen configured, personal email as test user

mcp-google-sheets added to claude_desktop_config.json — full absolute uvx path, OAuth credentials in env block

One-time browser auth flow — token saved automatically, no repeat login needed

06 · Tech stack

Claude Desktop mcp-google-sheets Google Sheets API Google Drive API OAuth 2.0 uvx Google Workspace Business Standard Native Drive connector — Docs only, not Sheets

07 · What I'd do differently

Honest reflection

Start with OAuth from day one — even without org policy restrictions, it's more portable and avoids the security risk of stored JSON key files. Treat Claude Desktop and claude.ai as completely separate systems from the start — different configs, different MCP connections, nothing shared. Always use absolute paths in Claude Desktop configs; "uvx" alone is a silent failure waiting to happen.