Google Drive MCP
There are at least 4 Google Drive MCP servers. Most guides only document one of them. Here's what nobody is explaining:
When developers search "google drive mcp server," they find GitHub READMEs, directory listings, and Google's docs — none of which compare the options or tell you which to pick.
Here's the actual landscape:
1. Official Anthropic reference server (`server-gdrive`)
Read-only. Exposes search, list, and read operations. Auto-converts Google Docs to Markdown and Sheets to CSV. Uses stdio transport — works with Claude Desktop, not Claude.ai web. Scope: drive.readonly. Best for read-and-analyze workflows.
2. felores/gdrive-mcp-server
Community TypeScript build with a tighter interface. Same read-only capabilities, better code structure if you want to extend it. Local stdio, Claude Desktop compatible.
3. michaelpine25/googleDriveMCP
Built specifically for Claude Desktop. Thorough setup documentation walking through GCP OAuth credential creation step by step. Good for first-timers.
4. Google Cloud managed MCP
Completely different category. Remote HTTPS endpoint hosted by Google. Full Drive tool surface: create, write, delete, move, share, manage permissions. IAM-controlled, audit-logged, shared drive support. Works with Claude.ai web. Requires GCP project setup.
The architectural distinction that matters most:
Community servers = local stdio = Claude Desktop only
Google managed = remote HTTP = Claude.ai web + team deployment
If you need write access, shared drives, or multiple users, the community servers are the wrong architecture entirely. If you just want Claude to read and analyze your Drive files locally, the reference server gets you there in 20 minutes.
The OAuth detail nobody mentions:
Your token.json file contains a refresh token that silently re-authenticates on every request. If you push a project containing it to a public GitHub repo, you've published live credentials with read access to your entire Drive.
Add token.json to .gitignore before your first commit. Non-negotiable.
Full comparison table (capability matrix, security breakdown, use case guide) is on MyMCPShelf → link in comments.
The MCP + Google Drive space is moving fast. Worth understanding which implementation fits your actual use case before you wire it into a production agent.
#MCP #ModelContextProtocol #GoogleDrive #AI #Claude #DeveloperTools #AIAgents