Google has recently released an open-source command-line interface (CLI) for Google Workspace, providing unified access to services like Gmail, Calendar, Drive, Sheets, Docs, Chat, Admin, and more.
This tool, often referred to as gws from the package Googleworkspace/cli, was announced and shipped in early March 2026. It’s explicitly designed for both human users; developers working in the terminal and AI agents, making it easier to build agentic workflows that interact with Workspace data without custom wrappers, browser automation, or third-party connectors like Zapier.
It pulls from Google’s Discovery Service at runtime, so commands auto-update when Google adds new API endpoints or methods. Human-friendly — Tab completion, –help per resource, –dry-run previews, auto-pagination, and easier OAuth handling than manual curl calls. Agent-ready — Outputs structured JSON by default, which LLMs and agents can reliably parse. Includes over 100 agent skills (SKILL.md files in the repo), covering individual APIs plus higher-level workflows and 50+ curated recipes for common tasks in Gmail, Drive, Calendar, Docs, Sheets, etc.
Then run commands like gws. Agent skills can be added directly, e.g., npx skills add github:googleworkspace/cli. Open-source under Apache 2.0. It’s described as experimental and community-oriented in some coverage; not a fully “official” Google product in the same way as gcloud, but introduced by Google Cloud figures and covered widely as a Google AI and Workspace release.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
This arrives amid growing interest in agentic AI that can read and write emails, manage events, edit docs and files, search Drive, etc. It simplifies integration for tools like Claude Code, Vertex AI agents, Gemini Enterprise agents, or custom LLMs that support shell execution or tool calling. The release has generated buzz in dev communities for reducing friction in automating Workspace tasks via AI.
Microsoft offers strong support for AI agents interacting with Microsoft Graph (the unified API for Microsoft 365 services like Outlook, Teams, OneDrive, SharePoint, Entra ID, Intune, and more), though the approach differs notably from Google’s Workspace CLI.
The official Microsoft Graph CLI often called mgc was a cross-platform. NET-based tool for calling Graph APIs from the terminal. However, Microsoft deprecated and archived it in late 2025, with full retirement scheduled around August 2026.
Reasons cited include declining usage and overlap with the more powerful Microsoft Graph PowerShell SDK using cmdlets like Invoke-MgGraphRequest or specific resource cmdlets. Community feedback, especially in AI/agent contexts, has highlighted this as a missed opportunity—Google’s agent-optimized Workspace CLI launched around the same time with features like structured JSON output, dynamic commands, and built-in agent skills/recipes.
For agentic workflows, LLMs like Gemini, Claude, or custom agents reading/writing emails, managing calendars, searching files, querying users/devices, etc., the ecosystem leans toward these:merill/msgraph Agent Skill (most direct analog to Google’s agent-ready CLI)Open-source Agent Skill (following the agentskills.io spec) that bundles the complete Microsoft Graph API surface (~27,700 endpoints) as local searchable indexes.
Agents get instant, offline knowledge of APIs, parameters, examples—no network calls for discovery. Optional direct execution: Authenticate once and let the agent call Graph APIs live supports GET/POST/PATCH/DELETE, but DELETE is blocked by default for safety.
Zero dependencies, cross-platform (macOS/Linux/Windows), installs via npx skills add merill/msgraph. Often paired with MCP (Microsoft Cloud Protocol) servers or tools like lokka.dev for execution. This is community-driven but widely adopted in agent builders for reliable, up-to-date Graph integration without wrappers.
Example from agent demos: Pipe mgc JSON outputs into processing scripts for compliance reports, user/device queries, etc. Agents compose pipelines autonomously, e.g. mgc devices list –filter “complianceState eq ‘noncompliant'” –output json | … Familiar CLI feel; works in terminals where agents have exec access.
Less “agent-native” than Google’s dynamic, JSON-first design—no built-in recipes/skills. Agent identities in Microsoft Entra ID (formerly Azure AD) for daemon/background agents calling Graph with application permissions. Emerging Microsoft 365 Agents Toolkit / Teams CLI integrations for building callable agents/functions in Teams/365.
Dynamic (runtime discovery), agent-first (JSON default, 100+ skills/recipes, tab completion + dry-run for humans too), open-source and actively positioned for AI agents. Microsoft’s side: Shifted to PowerShell SDK + community agent skills like msgraph. No official equivalent “agent-optimized CLI” yet, but the merill/msgraph skill fills much of the gap and is praised for speed/local search.
If you’re building agents that need to act on Microsoft 365 data, start with the msgraph Agent Skill—it’s the closest to a plug-and-play, agent-ready interface today. For production and enterprise, combine with official Entra agent auth + Semantic Kernel plugins. The space is evolving fast amid the agent boom, so watch for potential Microsoft announcements on native agent tooling.