Deadline: Published
The is_agentic flag: what to do before
Google added an "agentic" setting to OAuth clients so that Google and Workspace admins can tell AI-agent traffic apart from ordinary app traffic. Only clients marked as agentic can call the Google Workspace MCP servers.
Author: WorkspaceMCP.directory (Last verified )
Audience: Google Workspace Administrators & IAM Engineers · 3 min read
Key Actions Before
- : Setting available in Google Auth Platform > Clients.
- : Workspace MCP servers reject clients not marked as agentic ("the client isn't marked as an AI agent").
- : Setting locks on existing clients; afterwards you need a new OAuth client.
- Apps that only call Workspace REST APIs and aren't AI agents are not affected.
Google added an "agentic" setting to OAuth clients so that Google and Workspace admins can tell AI-agent traffic apart from ordinary app traffic. Only clients marked as agentic can call the Google Workspace MCP servers.
Key dates
| Date | What happens |
|---|---|
| Setting available in Google Auth Platform > Clients | |
| Workspace MCP servers reject clients not marked as agentic ("the client isn't marked as an AI agent") | |
| Setting locks on existing clients; afterwards you need a new OAuth client |
Checklist
-
List every OAuth client that calls
*mcp.googleapis.com(e.g.https://drivemcp.googleapis.com/mcp/v1) from Antigravity, Claude connectors, ADK, or Gemini Enterprise. - In Google Cloud Console → Google Auth Platform → Clients, open each one and select "Use this client for an AI-powered agent."
- If one client serves both agent features and non-agent features, create a separate client for the agent and mark only that one. Keep the existing client for non-agent use.
- Apps that only call Workspace REST APIs and aren't AI agents are not affected.
-
Missed the deadline? Create a new agentic client and update your config (e.g.
mcp_config.json).
Note for security MCP builders
The flag covers Google's MCP endpoints. Community admin/security servers (Alert Center, Directory) call REST APIs, often through a Domain-Wide Delegation service account. Keep those servers read-only by default and require human approval for remediation.
Sources
Need help securing Google Workspace for AI agents?
Talk to Revolgy →