Google Workspace MCP Server & CLI

Google Workspace MCP server and CLI for Claude Code, Codex and AI agents. 38 tools for Gmail, Drive, Sheets, Docs, Slides, Calendar, Tasks, Forms and Contacts through Google's official Workspace CLI, so your credential never leaves your machine.

One install gives you both surfaces, the same 38 tools under the same names, from the same server, so they cannot drift apart.

It wraps Google's own Workspace CLI, so your credential is created by you and stays on your machine. That also sidesteps the security assessment a third-party OAuth app needs for mailbox and Drive access.

Built by Navid Moazzez. Built on Slipway, which turns one definition of each tool into the MCP server and the CLI.

Two ways to use it

Command line

google-workspace-cli runs every tool as a command. Agents that run commands, like Claude Code, Codex and OpenCode, use it on their own, and you can type the same commands in a terminal, a script or a cron job:

google-workspace-cli                                          # every command, one line each
google-workspace-cli gmail-search --q "from:sarah newer_than:30d" --max 5
google-workspace-cli calendar-list-events --json
google-workspace-cli drive-search --q "name contains 'invoice'" --agent
google-workspace-cli gmail-create-draft --to a@example.com --subject Hi --body "See you Friday"
google-workspace-cli gmail-send-draft --draftId r-123 --confirm
google-workspace-cli which append rows to a sheet              # find the command for a task
google-workspace-cli <command> --help                         # what any command takes

--confirm is the shell spelling of the confirmation sending a draft needs. --json gives JSON, --compact puts it on one line, --select keeps only the fields you name, and --agent turns on all of it for a script. Exit codes are 0 ok, 1 unexpected, 2 usage or a refused write, 3 not found, 4 auth, 5 API, 7 rate limited and 10 nothing configured, so a script branches on the number.

google-workspace-cli schema <command> prints the exact JSON Schema an MCP client receives for that tool.

MCP server, for AI agents

google-workspace-mcp is what Claude Code, Claude Desktop, Cursor and the rest launch. You never run it by hand:

claude mcp add google-workspace -- npx -y @thenavidm/google-workspace-mcp-cli

In Claude Desktop, the .mcpb extension installs on a double click. Section 4 has every other client.

What each costs

Both surfaces are the same program with the same 38 tools. The difference is when the model pays for them. Measured in Claude Code:

Claude Code's tool search is on by default: it sends only the tool names and the server instructions, and loads a tool's full definition when the model reaches for it. An app that loads every tool up front pays the first line on every message, whether Google Workspace comes up or not. With the skill added, Claude Code also lists its one-line description, about 140 tokens.

To spend less, turn the server off when you are not using it, which in Claude Code is the /mcp panel. GWS_SERVICES=gmail,calendar lists only the services you name, and the rest cost nothing. Or install the CLI and add the server on the days it earns its place.

Measured on 2026-10-05 against 0.2.2, with Claude Code 2.1.286 on Claude Opus 5.5 (one short prompt with and without the server connected, once with ENABLE_TOOL_SEARCH=false and once with the default, the difference read from the API's own usage figures; SKILL.md the same way) and Codex 0.159.3 on gpt-6.1-sol, with a stand-in for the gws CLI so nothing reached a Google account:

The task was "find the command that appends rows to a Google Sheet, and the flags it requires". Every tool loaded costs less because each tool no longer repeats $schema and an execution block. Over the CLI, every 0.2.2 run read the general help, the command list and the command's help, three requests that each carry the conversation so far, and every 0.3.0 run read the general help and asked which, which answered with the command's help: two. Over MCP, Codex's model writes a script that prints either the tool it wants or the whole list, and it chose the whole list in five 0.3.0 runs and one 0.2.2 run of 10 each, before it had seen either version's tools; what each printout cost was the same or less for 0.3.0. Tool search costs the same within the spread of each version's own rounds, 803 to 810 tokens. Other apps and models count tokens a little differently, and tool-list characters divided by four are not API usage.

Related MCP server: Vopak Workspace MCP

Contents

1. What you can ask it 💬

• What did I agree to with the agency, and is it in the calendar?

• Summarize every unread email from this week and tell me which need a reply.

• Draft replies to the three that matter. Do not send them.

• Find the pricing spreadsheet and tell me what changed since March.

• Pull every response to the onboarding form into a new sheet.

• Who did I email about the launch, and did they reply?

• Build a doc from the notes in that folder.

• What is on my calendar next week that I could move?

The first one is the point. It reads mail and calendar together, which no single Google product does for you.

2. Quick install ⚡

Node 22 or newer. Nothing else.

npx -y @thenavidm/google-workspace-mcp-cli --version

That is the whole install. npx fetches it on demand, so there is nothing to update later.

For the CLI as a command you or your agent can run anywhere, install it once:

npm install -g @thenavidm/google-workspace-mcp-cli
google-workspace-cli

You also need Google's Workspace CLI, which is what actually talks to Google:

# macOS
brew install googleworkspace/tap/gws

# or download a binary
# https://github.com/googleworkspace/cli/releases

3. Setup 🔑

One command, once.

gws auth login

A browser opens, you approve, and you are done. The credential belongs to the CLI, not to this server.

Limit what it can touch while you are there:

gws auth login --services drive,gmail,calendar
gws auth login --readonly

You will see an "unverified app" warning

That is expected and it is not a problem. Google shows it for any OAuth app it has not reviewed, and reviewing an app that only you use is not worth the process. Click Advanced, then Go to (your app).

If you would rather not see it, gws auth setup walks you through creating your own Google Cloud project, and then the app is yours.

Revoking

gws auth logout

Or remove access at myaccount.google.com/permissions.

4. Connect your client 🔌

Claude Code

claude mcp add google-workspace -- npx -y @thenavidm/google-workspace-mcp-cli@latest

--scope user makes it available in every project rather than the current one.

Claude Desktop

The short way: download the .mcpb extension from the latest release and double-click it. It carries its own dependencies, so there is no config file to edit and nothing to install first. Install and sign in to gws first (section 3), because the extension talks to Google through it.

The long way, if you would rather edit the config yourself:

{
  "mcpServers": {
    "google-workspace": {
      "command": "npx",
      "args": ["-y", "@thenavidm/google-workspace-mcp-cli@latest"]
    }
  }
}

Tip Claude Desktop does not inherit your shell PATH, so it may not find npx or gws. If it fails, use absolute paths from which npx and which gws, and set GWS_BIN to the second one.

Quit Claude Desktop completely and reopen it.

claude.ai on the web

claude.ai runs connectors from Anthropic's cloud, not from your machine, so it cannot launch a local command. It needs this server running somewhere with a public HTTPS address. See section 7.

Cursor

.cursor/mcp.json

{
  "mcpServers": {
    "google-workspace": {
      "command": "npx",
      "args": ["-y", "@thenavidm/google-workspace-mcp-cli@latest"]
    }
  }
}

Windsurf

~/.codeium/windsurf/mcp_config.json, same shape as Cursor.

VS Code

.vscode/mcp.json. The key is servers, not mcpServers.

{
  "servers": {
    "google-workspace": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@thenavidm/google-workspace-mcp-cli@latest"]
    }
  }
}

Codex CLI

~/.codex/config.toml

[mcp_servers.google-workspace]
command = "npx"
args = ["-y", "@thenavidm/google-workspace-mcp-cli@latest"]

Everything else

Any stdio MCP client takes the same two things: the command npx and those arguments.

5. Check it worked 🩺

npx -y @thenavidm/google-workspace-mcp-cli doctor

It checks the CLI is present, that you are authenticated, and makes one live API call. It exits 0 when everything works and 1 when something does not. If every line is fine, restart your client and ask it what is on my calendar today.

6. Tools 🛠️

38 tools. The named ones cover what an assistant reaches for; workspace_raw covers the rest.

Gmail

Drive

Sheets, Docs, Slides

sheets_read · sheets_write · sheets_append · sheets_get · sheets_create docs_get · docs_create · docs_append slides_get · slides_create

Calendar and Tasks

calendar_list_events · calendar_create_event · calendar_update_event · calendar_delete_event · calendar_list_calendars tasks_lists · tasks_list · tasks_create · tasks_complete

Forms and Contacts

forms_get · forms_responses · contacts_search · contacts_list

Everything else

Read the schema before using workspace_raw, rather than guessing field names.

7. Running it on a server 🖥️

Needed for claude.ai, and useful if you want it always on.

curl -fsSL https://raw.githubusercontent.com/thenavidm/google-workspace-mcp-cli/main/deploy/install.sh | sudo bash

That creates a dedicated user, installs the gws binary with its checksum verified, generates a bearer token, and runs a systemd service bound to 127.0.0.1:8787. Nothing is exposed to the internet by the script: put it behind your existing reverse proxy, which is where TLS belongs.

Then authenticate as the service user. The script prints the exact command.

Docker instead:

docker build -t google-workspace-mcp .
docker run -d --name gws-mcp -p 127.0.0.1:8787:8787 \
  -e GWS_MCP_TOKEN="$(openssl rand -hex 32)" \
  -v ~/.config/gws:/home/node/.config/gws \
  google-workspace-mcp

The config mount must be read-write: gws stores its encryption key and cached discovery documents there, not just a token.

Note On macOS the credential lives in the OS keyring, so copying ~/.config/gws to a Linux box does not work. Set GOOGLE_WORKSPACE_CLI_KEYRING_BACKEND=file and authenticate on the server itself.

8. Safety 🛡️

Email is drafted, never sent silently. gmail_create_draft writes to Drafts and stops. Sending is a separate tool that needs confirming, because a sent email cannot be recalled.

Confirmation on what cannot be undone: gmail_send_draft, drive_trash, calendar_delete_event, and workspace_raw when its method deletes, removes, trashes or clears. Over MCP a person approves each call: Claude Code (2.1.246 and later) shows its own prompt, and a client that can show forms asks with an approval form whose one box starts unticked. Where a client can do neither, the model's confirm: true still counts, and GWS_CONFIRM=model makes it enough everywhere. In a terminal the flag is --confirm, which --agent never adds.

Nothing is deleted permanently. drive_trash is recoverable for 30 days. There is no hard-delete tool.

Turn writes off, or narrow the surface:

GWS_READ_ONLY=1                  # hide and refuse every write; workspace_raw stays for reads
GWS_SERVICES=drive,calendar      # nothing else is even listed
GWS_AUDIT_LOG=~/gws-audit.jsonl  # one line per attempted write, with who approved it

GWS_SERVICES removes the other tools from the list rather than failing when called. A model cannot reach for a tool it cannot see.

Settings

The program reads the environment directly. It does not load .env files. The Google credential is the gws CLI's, so none is set here.

Prompt injection. Anything read from a mailbox or a shared document was written by someone else and can contain text shaped like an instruction. The server tells the model to treat it as data. That helps and is not a guarantee: for an agent working unattended on other people's content, GWS_READ_ONLY=1 is the real defence.

Full detail in SECURITY.md.

9. Troubleshooting 🔧

Run doctor first. It answers most of it.

10. FAQ

An MCP server is a standard way to give an AI assistant real access to a tool, so it can act rather than guess. You install it once, your assistant gains a set of tools, and the same server works in Claude, Cursor, ChatGPT and anything else that speaks MCP.

Without one, an assistant can only talk about your email. With one, it can read it.

google-workspace-cli is the same program as the MCP server, run as commands. AI agents that run commands, like Claude Code, Codex and OpenCode, use it on their own, and you can type the same commands in a terminal, a script or a cron job. Every tool is a command with dashes, so gmail_search runs as google-workspace-cli gmail-search.

Use the MCP server in an app with no terminal, like Claude Desktop's chat. Use the CLI anywhere commands run: an agent like Claude Code, Codex or OpenCode, a script or a cron job. The MCP server's tools take up context on every message, and the CLI costs nothing until it runs.

Google's suite of work apps: Gmail, Drive, Docs, Sheets, Slides, Calendar, Tasks, Forms and Contacts. If you use a Gmail address, you already have it.

You need to be comfortable running two commands in a terminal: one to install, one to sign in. After that it is all conversation with your assistant.

If a terminal is unfamiliar, the Claude Desktop path in section 4 is the shortest route.

Your Google data goes from Google to the machine running this server, and from there to whichever AI client you connected. It does not pass through any server of ours, because there isn't one.

The exception is deliberate and worth understanding: whatever your assistant reads, it sends to its own model provider, the same as anything else you paste into a chat.

Work across apps in one step. Gmail cannot search your Drive, and Calendar cannot read your mail. An assistant with this server does both in a single question, and it can act on the answer: draft the reply, create the event, update the sheet.

It can trash a Drive file and delete a calendar event, and both need confirming: over MCP a person approves each in the client, and in a terminal it takes --confirm. Trashed files are recoverable for 30 days. There is no permanent-delete tool.

It cannot send an email by accident: drafting and sending are separate tools, and sending needs confirmation.

If you want none of that, run it with GWS_READ_ONLY=1.

It costs nothing. The server is free and open source, Google's CLI is free, and the Workspace APIs are free at any volume a person generates.

It works with any MCP client. Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Codex CLI and Gemini CLI all run it locally.

claude.ai is the exception: it connects from Anthropic's cloud rather than your machine, so it needs the server running somewhere with a public HTTPS address. See section 7.

One account per server. The gws CLI holds a single login.

For a second account, run a second copy with GOOGLE_WORKSPACE_CLI_CONFIG_DIR pointing somewhere else, and add it to your client under a different name.

It refreshes itself. The CLI holds a refresh token and renews access silently, so you sign in once.

If you ever do need to sign in again, doctor says so plainly rather than failing with a permissions error.

Questions

Run into a problem or have a question? Open an issue and I will help.

Built by

Navid Moazzez

Navid Moazzez is a leading AI business strategist, and the host of the AI Creator Summit, watched by 100,000+ creators. He helps creators and founders master AI and build their own AI Operating System (AI OS) to automate their business and life. He creates useful free tools, MCP servers and CLIs that creators and founders can use in their own workflows.

Dependencies

License

MIT. Free to use, modify, and share.

Not affiliated with, endorsed by, or sponsored by Google LLC. Google Workspace, Gmail, Google Drive, Google Docs, Google Sheets, Google Slides, Google Calendar, Google Tasks, Google Forms and Google Contacts are trademarks of Google LLC. This project wraps the official Google Workspace CLI and holds no credentials of its own.

© 2026 Navid Media. Made with ❤️ by Navid Moazzez.