This article is for client-side IT administrators, technical sponsors, and Gemini Enterprise administrators enabling PI's MCP connector during the current pre-beta phase.
Learn how to enable the Predictive Index MCP connector, and get PI-powered behavioral intelligence right in Google Gemini.
Tip: Before you start, make sure you've met the PI-specific prerequisites for enabling the MCP.
Note: The Predictive Index MCP is currently in pre-beta. We plan to make regular changes and updates on the road to its full release. Google's support for custom MCP servers in Gemini is also in preview, so its menu labels may change.
Who can add the connector to Gemini?
Gemini Enterprise, Business edition: A team administrator adds the connector once for the team. Team members then use it with their own PI accounts.
Gemini Enterprise, Standard and Plus editions: A Google Cloud user with the Discovery Engine Editor role adds the connector to your Gemini Enterprise app. Users then connect their own PI accounts.
Gemini Enterprise connects to PI from Google's cloud, not from your device. For details, see The Predictive Index MCP: Security, privacy, and IT reference.
Note: This article covers Gemini Enterprise. For the Gemini app, see the FAQs below. Gemini CLI is not covered here.
What you'll need
Field | Value |
Name | Predictive Index |
Server URL | |
Authentication | OAuth 2.0 |
Authorization URL | |
Token URL | |
Auth URL Parameter | Leave blank |
Client ID | ez72AHJZIsjaxciDam1Hb1heGNNwdhq3 |
Client Secret | Any value, for example not-used |
Scopes | openid profile email offline_access |
Enable PKCE Support (Standard and Plus only) | Selected |
Note: Gemini Enterprise doesn't register with PI automatically, so you enter these values by hand. The client ID is a public identifier, not a password. PI doesn't issue a client secret and ignores any value you enter, but Gemini requires the field. PI already allows Gemini Enterprise's redirect URL, so you can skip Google's step to register it.
How to connect PI to Gemini Enterprise: Business edition
Important: You must be a team administrator in Gemini Enterprise, with a PI account that has Obi enabled, to perform these steps.
Go to business.gemini.google and sign in as a team administrator.
Open your team's settings (the gear icon next to your name), then select "Connected apps."
Click "Add MCP Server."
Enter the Name and Server URL from the What you'll need table above, then continue.
Choose "OAuth 2.0" and enter the Authorization URL, Token URL, Client ID, Client Secret, and Scopes from the same table. Leave "Auth URL Parameter" blank.
Click "Add." When prompted, sign in with your PI account and approve.
In the Connected apps list, make sure "Enabled for team" is turned on for Predictive Index.
Ask team members to complete "Connecting your account" below.
Note: If you remove the connector and add it again, team members will need to reconnect.
How to connect PI to Gemini Enterprise: Standard and Plus editions
Important: You need the Discovery Engine Editor role in the Google Cloud project that hosts your Gemini Enterprise app, and a PI account with Obi enabled. Your organization policy must also allow custom MCP data stores, and if you restrict outbound traffic, it must allow mcp.predictiveindex.com.
In the Google Cloud console, open the Gemini Enterprise page, go to Apps, and select your app.
Go to Connected Data Stores and click "New data store."
Under Select a data source, search for "Custom MCP Server" and click "Add MCP server."
Leave "Route egress through Agent Gateway" cleared.
Choose "OAuth 2.0" and enter the values from the What you'll need table above. Make sure "Enable PKCE Support" is selected.
Click "Verify Auth," sign in with your PI account, and approve. Then click "Continue."
Skip Advanced options and click "Continue."
Pick a location, name the connector Predictive Index, and click "Create." Wait until its state changes from "Creating" to "Active."
Open the new data store, go to Actions, and click "Reload custom actions." Select all of PI's tools and click "Enable actions."
Ask users to complete "Connecting your account" below.
Note: PI marks its tools as read-only, so Gemini can use them without asking for confirmation each time. For Google's full steps, see: Set up your custom MCP server data store.
Connecting your account (each user)
Open Gemini Enterprise and start a new chat.
Ask a question that needs PI, for example: "Ask Obi what my behavioral profile is." If Predictive Index isn't available, your administrator hasn't added it yet or hasn't turned it on.
If Gemini asks you to authorize Predictive Index, sign in with your PI account and approve. You'll return to Gemini, connected.
Run the test in "Verifying PI's MCP connection."
Tip: Once the MCP is enabled, learn how to get the most out of it here.
FAQs
I don't see "Predictive Index" in Gemini Enterprise. What now?
I don't see "Predictive Index" in Gemini Enterprise. What now?
An administrator has to add the connector first. Ask your Gemini Enterprise administrator to check the following based on your plan:
In Business edition, "Enabled for team" must be on.
In Standard and Plus, the data store must show "Active" and its actions must be enabled.
Sign-in fails in Gemini Enterprise. What should I check?
Sign-in fails in Gemini Enterprise. What should I check?
Check that the Authorization URL, Token URL, Client ID, and Scopes exactly match the What you'll need table above. If they match and sign-in still fails, remove the connector and add it again.
Can I add PI to the Gemini app at gemini.google.com?
Can I add PI to the Gemini app at gemini.google.com?
Not with a work account. Google allows custom apps only on personal Google accounts. With a work account, Gemini may let you add PI, but it doesn't use PI's tools in chat. Use Gemini Enterprise instead.
The UI steps for Gemini are inaccurate or outdated. What now?
The UI steps for Gemini are inaccurate or outdated. What now?
Google's menu labels and navigation flows change over time. If a label above doesn't match what you see, use our host-agnostic steps for general guidance.
If you need more help configuring our MCP, please contact our Product Manager, Stefano Hernandez, at [email protected].
Additional support
Setting up PI's MCP connector (host-agnostic steps)