TL;DR
Google released GWS CLI — a command-line tool to control Drive, Sheets, Gmail, Calendar, and more. This means you no longer need MCP servers to connect AI agents to Google Workspace. The real solution to "AI needs to talk to services" isn't a middleware protocol — it's services shipping their own CLIs.
The Problem MCP Tried to Solve
MCP (Model Context Protocol) was a clever idea. AI models had no standardized way to talk to external services, so MCP proposed a universal adapter layer.
Want your AI to access GitHub? Spin up an MCP server. Need Slack integration? Another MCP server. Notion? You guessed it — yet another MCP server.
The architecture looks like this:
AI Agent → MCP Server → Actual Service API
An extra translation layer in the middle. And that's exactly where the problems start.
The fundamental limitations
- Complex setup: You either build MCP servers yourself or trust third-party ones with your data
- Maintenance overhead: Every time a service API changes, the MCP server needs updating too
- Unnecessary middleware: If an AI can call a CLI directly, why do we need a server in between?
- Security concerns: Your data flows through an intermediary you may not fully control
Here's the core issue: MCP tried to solve "AI has no way to reach services," but the real answer is services providing AI-friendly CLIs themselves.
Google GWS just proved it.
Google GWS CLI: A Paradigm Shift
Look at these commands:
# List files in Google Drive
gws drive files list --params '{"pageSize": 10}'
# Read spreadsheet data
gws sheets spreadsheets values get \
--params '{"spreadsheetId": "ID", "range": "Sheet1!A1:C10"}'
# Search Gmail
gws gmail messages list --params '{"q": "from:boss subject:urgent"}'
# Send a message in Google Chat
gws chat spaces messages create \
--params '{"parent": "spaces/SPACE_ID"}' \
--json '{"text": "Deploy complete"}'
This isn't just a developer convenience tool. This is Google's bet on the future.
What Google is really saying: "The next primary user of Google Workspace might not be a human."
Think about it. Who will be opening Chrome, navigating to drive.google.com, and clicking through folders? A person? Or an AI agent like Claude Code?
GWS CLI is an interface designed for AI agents. No browser GUI needed. JSON in, JSON out. That's what a CLI is.
Now there's zero reason to set up a Google Drive MCP server. Just call gws drive files list directly.
The Terminal Is Back (But Not for Humans This Time)
Quick history lesson:
1970s: Everything was terminal. Linux, Unix, command line. Creating files, sending emails, querying databases — all terminal.
1990s: GUIs and web browsers arrived. Terminal became "developer-only territory." Regular users moved to mouse-and-click.
2024-2026: We're going back to terminal. But this time, AI agents are the ones typing commands.
Why terminal is perfect for agents
Browser GUIs were designed for human eyes and hands — colors, buttons, icons, animations. For an AI agent, all of that is noise.
CLI deals in structured I/O only. Command in, JSON out. Error happens, error code returned. It's the perfect format for AI to consume.
Where does Claude Code live right now? In the terminal. The hub of the agentic OS is already the terminal.
What I'm Actually Doing in Terminal Right Now
I'm writing this post from Claude Code in a terminal. Not in a browser, not in Google Docs, not in Notion.
Here's what my workflow looks like — all terminal, zero browser tabs:
# Content creation
claude "Write a blog post about MCP vs CLI"
# Google Drive file management
gws drive files list --params '{"q": "name contains '\''report'\''", "pageSize": 10}'
# Spreadsheet data processing
gws sheets spreadsheets values get \
--params '{"spreadsheetId": "abc123", "range": "Sheet1!A1:D20"}'
# Email handling
gws gmail messages list --params '{"maxResults": 5}'
# Social media publishing
python3 scripts/publish_to_social.py --platform threads
# Image generation
python3 scripts/generate_image.py "prompt" -o output.png
# Code deployment
git push origin main # CI/CD handles the rest
How many of these require opening a browser? Zero.
What Survives After MCP
MCP dying doesn't mean AI stops connecting to external services. It means the connections get more direct and simpler.
What stays
1. Official CLI tools — Like GWS. The service itself provides a CLI. No middleware. AI calls it directly.
# Google → gws
gws drive files list
# GitHub → gh
gh issue list --repo owner/repo
# Vercel → vercel
vercel deploy --prod
# Stripe → stripe
stripe payments list --limit 10
2. Direct API calls — For services without CLIs, curl or a Python script hits the REST API directly. No MCP server needed.
# Direct API call — no MCP server required
curl -s -X POST https://api.example.com/data \
-H "Authorization: Bearer $API_TOKEN" \
-H "Content-Type: application/json" \
-d '{"query": "latest metrics"}'
3. Agentic OS native tools — Claude Code, Gemini CLI, and similar tools that are terminal-first by design.
What dies
- Third-party MCP server ecosystems
- Browser-based AI integration tools
- Complex middleware protocols
FAQ
Should I abandon my MCP projects right now?
Not yet. MCP is still the only AI integration method for many services. But the direction is clear. For new projects, think twice before building a new MCP server. If an official CLI or direct API call is available, choose that path — it's better long-term.
Do non-developers need to learn terminal?
Here's the paradox: the terminal is returning, but humans don't need to memorize commands this time. AI agents type them for you. A non-developer can say "Find this month's reports on Google Drive" and the agent runs gws drive files list behind the scenes.
The key isn't knowing terminal commands — it's understanding what terminal can do.
Are other services besides Google shipping CLIs?
Many already do: Stripe CLI, GitHub CLI (gh), Vercel CLI, Shopify CLI, and more. GWS will accelerate this trend — "Google did it, so we should too" pressure will spread across the industry. Expect most major SaaS services to ship official CLIs within 1-2 years.
What's your take — are you seeing MCP usage decline in your own workflows? Have you started replacing MCP servers with direct CLI calls? Drop a comment below.
Top comments (0)