TL;dr: On October 12, Google announced that OAuth clients used by AI agents to call Google Workspace MCP servers must be marked as agentic, and starting November 11 the Workspace MCP servers for Gmail, Drive, Sheets, Calendar, Chat, Docs, and Slides will reject any client that is not marked. If you have an AI agent that reads or writes your spreadsheets through one of these integrations, you have until November 11 to open the Google Cloud console, go to Google Auth Platform > Clients, and check "Use this client for an AI-powered agent" on the client your agent uses. Miss the window and the setting locks on November 12, which means creating a brand new OAuth client and rewiring the integration.

Today, we will be looking at what is actually changing, why Google is doing it, and the three checks I would run on my own setup before the deadline. This matters to you even if you have never opened the Cloud console, because the agents people are wiring into their finance workflows, the n8n flows, the Apps Script helpers, the MCP connectors, all authenticate through OAuth clients, and Google just drew a line between human traffic and agent traffic.

What changes on November 11 is narrow but strict. Google is rolling out Workspace MCP servers, which expose Gmail, Drive, Sheets, Calendar, and the rest as Model Context Protocol servers that external AI agents can call. Only OAuth clients with the agentic flag set will be allowed to call them. If your agent calls the regular REST APIs directly, nothing changes for you, and Google says so explicitly. The enforcement is only at the MCP layer. But if your agent goes through an MCP server, and more of them will over time, an unmarked client gets rejected with an error saying the client is not marked as an AI agent.

The reason Google treats agents differently from apps comes down to auditability, and this is the part worth sitting with. A regular app does what its code says, and its code changes when someone ships a release. An agent decides what to do at runtime, which means the same OAuth scopes can produce very different behavior from one run to the next. Google is not granting agentic clients any extra scopes and it is not bypassing user consent, which is worth stating plainly because that rumor is already circulating. What the flag does is isolate agent traffic so Workspace administrators can see it separately and apply dedicated policies to it without touching the humans. It is a name badge rather than a hall pass.

Here are the three checks I would run. First, inventory every OAuth client that touches your Google data. Open the Cloud console, go to Google Auth Platform > Clients, and list them. For each one, write down which integration uses it, which scopes it holds, and whether an AI agent is behind it. Most people will find clients they forgot about, which is exactly the point of the exercise. Second, for every client an agent uses, check the "Use this client for an AI-powered agent" box before November 11, and remember that on November 12 the setting locks permanently on existing clients, so there is no fixing it later without rebuilding the credential. Third, split any client that serves both humans and agents. Google's own guidance says to create a separate OAuth client for the AI agent and keep the existing one for everything else, because once agent traffic is tagged, your admin can put agent-specific controls on it without breaking the human workflows that share the credential today.

The rule I am adopting for my own setup is simple. No agent writes to a model it cannot audit. That means every agent with write access to a spreadsheet gets a row in an inventory with its client, its scopes, its owner, and its risk tier, plus a log of exactly what it is allowed to write and who approved it. If a write is not on the list, the agent should not be doing it. This is the same discipline as a close checklist, applied to credentials instead of accounts. It takes an hour to set up and it answers the only question that matters when something goes wrong, which is what had access to what.

Before the deadline, do the inventory, flip the flag on your agent clients, and split the shared ones. The template below has both tabs ready with example rows so you can see the shape of it. Copy it, fill in your own clients, and you will know exactly where you stand before November 11 rather than finding out from an error message on November 12.

Template: template: AI-agent access audit (Google Sheets, make a copy to use it)

- Francois