EMPVResearch NotesIT
EMPV / RESEARCH NOTESNOTE / 018

NEWS NOTE / AI INTEGRATION

Google Workspace MCP: AI agents face new OAuth requirements from November 11

Google is introducing an OAuth 2.0 client requirement for AI agents using Workspace MCP servers. Traditional REST API integrations are not affected.

Published2026-11-02 · 3 min
Google WorkspaceModel Context ProtocolOAuth 2.0AI AgentsSecurity
01 / The change

A new requirement for AI integrations

Starting November 11, 2026, applications using Google Workspace Model Context Protocol (MCP) servers will need OAuth 2.0 clients explicitly identified as being used by AI-powered agents.

Google announced the change on October 12. It applies to MCP servers for Gmail, Google Drive, Calendar, Docs, Sheets, Slides and Chat.

MCP allows AI assistants to interact with external tools, retrieve information and perform supported actions. In Workspace, these tools can help an agent search documents, access information or interact with services within the permissions granted to the application.

Under the new requirement, MCP servers will reject requests from OAuth clients that are not marked as agentic.

02 / Authentication

What changes for OAuth clients

Google Auth Platform includes a setting called «Use this client for an AI-powered agent». Developers calling Google Workspace MCP servers must enable this setting for the corresponding OAuth 2.0 client.

Google also specifies a second date. From November 12, clients that remain unmarked will be treated as non-agentic and will not be able to access Workspace MCP servers. Applications will need a newly created agentic OAuth client to restore that access.

This requirement does not apply to every integration using Google Workspace. For example, an application accessing Google Drive exclusively through standard REST APIs is not affected.

Where the same OAuth client serves both traditional application features and AI agents using MCP, Google recommends separating those functions into different clients.

03 / Checks

What developers should check

Teams operating Workspace-connected AI agents should first identify which integrations actually call Google's MCP servers.

They should then inspect the associated OAuth clients in Google Auth Platform and configure dedicated agentic clients where required.

Testing existing workflows is also important, particularly where an agent retrieves information or performs authorized operations across multiple Workspace services.

There are documented exceptions. According to Google's configuration guidance, service accounts using domain-wide delegation are not rejected for lacking this specific setting. Their access remains governed by the administrator's delegation configuration.

The Workspace MCP servers covered by the announcement are currently available through Google's Developer Preview Program.

04 / Implications

Agent identity becomes part of integration design

The change introduces an explicit distinction between an application calling an API and an AI agent accessing tools through MCP.

Marking an OAuth client as agentic does not, by itself, establish which company data an agent may access or which actions it may perform. Those boundaries still depend on authorization and application-level controls.

As AI agents become connected to internal documents, communications and business systems, their technical identity becomes another part of integration design.

Our earlier Research Note on ChatGPT, ERP integrations and MCP covers the related distinction between technical tool access and the permissions required to use it.

EMPV / TAKEAWAY

From November 11, agents using Google Workspace MCP servers must also be identified in their OAuth configuration.

EMPV / SHARE

Share this Research Note.

Research Notes