A new command-line tool published to GitHub consolidates Workspace’s sprawling APIs into a single interface. It also signals how seriously the company is taking the agentic AI moment.
The tool, whose documentation describes it as “one CLI for all of Google Workspace, built for humans and AI agents,” is called gws. It provides unified command-line access to Gmail, Google Drive, Calendar, Docs, Sheets, Slides, Chat, and most other Workspace services.
But the more revealing detail is buried in the instructions: the documentation includes a dedicated integration guide for OpenClaw, the open-source AI agent that went viral in late January and has since become something of a Rorschach test for where agentic AI is headed.
Google’s decision to name-check OpenClaw in official documentation, even unofficial official documentation, is not something companies do by accident.
Why a command-line tool matters for AI agents
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
Before GWS, an AI agent that wanted to search a Gmail inbox, pull a file from Drive, and update a Calendar event had to navigate three separate APIs, each with its own authentication flows, rate limits, and response formats. The process worked, but as PCWorld described it, it was “a royal pain.”
The new tool collapses that into a single interface. Every operation produces structured JSON output the format AI agents can parse reliably without the ambiguity that can derail graphical interfaces. Authentication is handled once via OAuth, then inherited by any agent that calls the tool.
The architecture has one particularly elegant feature: gws does not ship a static list of commands. Instead, it reads Google’s own Discovery Service at runtime and builds its entire command surface dynamically. When Google adds a new API endpoint, the tool picks it up automatically.
There is no version to update, no stale documentation to wrestle with. For agents designed to work across long time horizons, that self-updating quality is not a minor convenience; it is a meaningful reliability guarantee.
The repository also includes more than 100 pre-built “agent skills” covering common Workspace workflows: uploading files to Drive with automatic metadata, appending data to Sheets, scheduling Calendar events, forwarding Gmail attachments, and dozens of similar operations.
These are the discrete, composable building blocks that agent frameworks like OpenClaw are designed to chain together.
The OpenClaw connection
OpenClaw’s story has moved fast. The project was published in November 2025 by Peter Steinberger, an Austrian software developer, under the name Clawdbot, a name that drew a trademark complaint from Anthropic.
After a brief stint as Moltbot, it settled on OpenClaw in late January 2026. Within weeks, users had created 1.5 million agents using the platform; the GitHub repository accumulated nearly 200,000 stars. OpenClaw’s premise is simple enough to fit on a business card: AI that actually does things.
On 14 February, Sam Altman announced that Steinberger was joining OpenAI to lead the next generation of personal agents. OpenClaw would move into an independent open-source foundation that OpenAI would support. “The lobster is taking over the world,” Steinberger wrote in his farewell post. “My next mission is to build an agent that even my mum can use.”
Google’s Workspace CLI landing in the middle of that story, with OpenClaw integration instructions in the documentation, three weeks after Steinberger joined OpenAI, is the kind of timing that does not look accidental. Whether it reflects a deliberate competitive response, a coincidental release, or simply developers at Google shipping something that was already in progress is not confirmed.
What is clear is that a major platform company has now built infrastructure specifically to make its apps more useful for the open-source agent ecosystem that OpenAI just acquired the architect of.
MCP and the broader picture
Beyond OpenClaw, gws also functions as a Model Context Protocol server. MCP is the open standard for how AI agents communicate with external tools, originally developed by Anthropic and now adopted across the industry. Running gws mcp exposes Workspace APIs as structured tools that any MCP-compatible client, Claude Desktop, VS Code with AI extensions, or Google’s own Gemini CLI, can natively call.
That MCP support is significant because it means the tool is not merely an OpenClaw utility. It is infrastructure for the entire class of AI agents that is converging on MCP as a standard. Google is, in effect, making Workspace a first-class citizen in the emerging agent ecosystem, regardless of which model or framework is doing the work.
One important caveat: Google’s documentation explicitly notes that gws is “not an officially supported Google product.” It is published as a developer sample, meaning there are no guarantees of stability, security, or ongoing maintenance at the level of a production service. For individual developers and experimenters, that is a manageable risk.
For enterprises considering deploying AI agents against live Workspace data, it is a meaningful limitation, particularly given the ongoing concerns about OpenClaw’s security model, which a Cisco research team found vulnerable to data exfiltration and prompt injection via malicious third-party skills.
What Google is signalling
Addy Osmani, Director of Google Cloud AI, has framed his team’s focus as building infrastructure for agentic systems, those capable of generating command-line inputs and managing structured outputs across complex workflows. The Workspace CLI fits that vision directly.
The broader pattern is legible. Microsoft has Copilot Tasks. OpenAI now has the architect of OpenClaw. Google has its own Gemini agent stack, and now a CLI that makes its most widely-used productivity suite readable by any agent that speaks JSON and MCP.
The competition for where enterprise AI agents live and what data they can reach is accelerating, and the battleground increasingly looks like the infrastructure beneath the applications, not the applications themselves.
For now, gws is a GitHub repository with a caveat. But the 14,000 stars it accumulated before most journalists noticed suggest that developers who build agents for a living already understand what it means.
