Connect Gemini Enterprise Standard/Plus to Agent Authority
Add SecureAuth Agent Authority to Gemini Enterprise as a Custom MCP Server so every tool call is checked and audited
When you finish, every MCP tool call Gemini Enterprise makes is checked against your policies and recorded in the audit log. Google's Gemini Enterprise (Standard or Plus edition) reaches your approved tools through SecureAuth Agent Authority, added as a data store in the Google Cloud console, and each user signs in as themselves.
Before you begin
- Gemini Enterprise Standard or Plus edition – the editions this setup supports.
- Google Cloud project with billing enabled – with the AI Applications (Discovery Engine) API turned on.
- Organization Policy Administrator role (
roles/orgpolicy.policyAdmin) – granted at the Google Cloud organization level, to allow custom MCP connectors. Having the Organization Administrator or Owner role isn't enough: neither one includes this permission. - Administrator access to Agent Authority – in your Agentic AI workspace, to check Agent Access, and to change a rule if one blocks Gemini Enterprise.
- Policy that allows Gemini Enterprise's tool calls – see Confirm a policy allows the agent.
Administrator and user sign-ins
Setup involves two separate sign-ins:
- Administrator setup – an administrator clicks Verify Auth and signs in once while setting up the data store, which proves the client values are correct.
- Per-user consent – each user clicks Authorize and signs in once in the Gemini Enterprise web app before their own tool calls run.
Create the OAuth client in Agent Authority
An OAuth client is the credential that lets Gemini Enterprise sign your users in to Agent Authority. You create it here and paste its values into the Google Cloud console, so keep Agent Authority and the console open in separate tabs.
-
Open the connect flow.
In Agent Authority, go to Agents & NHIs, click Connect, then choose Gemini Enterprise.
Connect your agent → Gemini Enterprise walks you through creating the OAuth client and shows what to paste into Google Cloud. -
Name it and create.
Enter a name and click Create client. Use a distinct name for each Google Cloud environment you connect. Leave the redirect URI as shown: Google fixes it athttps://vertexaisearch.cloud.google.com/oauth-redirect. -
Copy the values now.
The client secret is shown only once
If you lose it, delete the agent and reconnect to issue a new client.

After Create client, Agent Authority reveals the values in Gemini's field order. Copy each one, then paste it into the Custom MCP Server connector. You paste them in Where each value goes.
Create the data store in Google Cloud
-
In the Google Cloud console, search for AI Applications.
-
Go to Data Stores → Create data store, search for
MCP, and choose Custom MCP Server.
Search for custom MCP on the Source step to find the Custom MCP Server tile. -
Paste each value from Agent Authority into the matching field, as listed in Where each value goes.
Paste them rather than retyping: the gateway URL contains an identifier unique to the organization.
The MCP Server Configuration form with each value pasted from Agent Authority.
Where each value goes
Match each field by its label, not its position: Google changes this form's layout from time to time.
| Connector field | Paste from Agent Authority |
|---|---|
| MCP Server URL | MCP Server URL (HTTPS, StreamableHTTP transport) |
| Authorization URL | Authorization URL |
| Authorization URL Parameters | Authorization URL Parameters (&access_type=offline&prompt=consent, for refresh tokens) |
| Token URL | Token URL |
| Client ID | Client ID |
| Client Secret | Client Secret |
| Scopes | Scopes, unchanged |
Set the authentication options
On the same form, set:
- Select your authentication method – OAuth 2.0.
- Enable PKCE Support – off, because the gateway issues a client secret.
- Use HTTP Basic Authentication – off, because the gateway expects the client secret in the request body.

Stay under Gemini's tool limit
Gemini Enterprise accepts at most 150 tools from one connector.
For more, turn on tool search: add ?tools=search to the end of the MCP Server URL.
If the URL already contains a ?, add &tools=search instead.
Gemini then sees only search_tools and invoke_tool, and reaches every allowed tool through them.
Tool search needs MCP Gateway › Tool search set to Opt-in (the default) or On.
Connecting to just one MCP? To stay under an agent's tool limit, point the URL at a single MCP —
/gateway/mcp/{slug}instead of/gateway/mcp. Grab the exact URL from the resource's detail page; see Scope a connection.
Verify the connection
- Click Verify Auth at the bottom of the form.
- Sign in through SecureAuth and approve the consent prompt.
Successfully logged in means the client ID, client secret, and URLs are correct. Fix any error before you continue to Configuration: Google doesn't create the data store until Verify Auth succeeds.
Finish the configuration
- Region – choose global. It can't be changed later.
- Sensitive Data Protection – skip this step unless the organization already has a policy, then click Continue.
Allow custom MCP connectors
Google Cloud blocks custom MCP connectors by default.
At the Pricing step, the console may show Operation denied by org policy, naming the constraint constraints/discoveryengine.managed.disableCustomMcpServerConnector.
To allow the connector, override this Google Cloud organization policy on the data store's project. You need a role granted at the organization level, but the override applies only to that project, and the rest of the organization stays blocked.
-
Go to IAM & Admin → Resource Management → Organization Policies.
-
In the resource picker, select the data store's project, so the change applies to that project only.
-
Find Disable custom MCP server connector for Gemini Enterprise and click Manage policy.
Manage policy needs the Organization Policy Administrator role
Without it, the button is unavailable. Having the Organization Administrator or Owner role isn't enough: neither one includes this permission. Google grants this role for the whole organization. To limit it to certain projects, add an IAM condition to the role grant.
-
Select Override parent's policy, add a rule with Enforcement: Off, click Done, then Set policy.

Once the override is set, the policy shows Inactive with the source Override parent's policy.
Choose the identity provider
Google Cloud prompts you to configure an identity provider, which is how Gemini Enterprise identifies each user. Under AI Applications → Settings → Authentication, select Google Identity and save.

Confirm a policy allows the agent
Gemini Enterprise sees only the tools a policy allows, so check before you enable tools.
-
In Agent Authority, go to Agent Access.
-
Check that an Active rule allows Gemini Enterprise. The Allow all rule that new organizations start with does. Draft and disabled rules aren't enforced.
Rules are evaluated top to bottom, and the first match wins. A tool call that no rule matches is denied.
If no rule matches, create a policy first.
Enable tools
Google lists the gateway's tools as actions.
- In AI Applications → Data Stores, open the new data store.
- Open Actions → Reload custom actions.
- Select the tools to make available and click Enable actions.
The list shows only the tools a policy allows.
With tool search on, it shows only
search_toolsandinvoke_tool.
Check that the connector and action states both show Active.
Create the Gemini Enterprise app
-
Go to AI Applications → Apps → Create app.
-
Under Search and assistant, click Learn More and select Gemini Enterprise.

The Gemini Enterprise tile is on the Search and assistant tab. -
Name the app and keep the region set to global.
-
Under Advanced Options, add your company name, which improves response quality.
-
Click Create.
-
Open Connected data stores and attach the new data store.
Check it works
-
Open the app's web app URL from its Overview page and sign in.
-
Click the Connectors icon for the data store and click Authorize. Each user does this once.

Each user clicks Authorize next to the data store in the connectors menu. -
Ask a question that needs one of the gateway's tools.
Name the MCP connection in your request
Gemini may not use the MCP data store on its own: it can answer from a web
search or say the integration isn't connected. Name it in your request, such
as use the MCP connection to list my Jira tickets.
- In Agent Authority, go to Agent Trail and check that the tool call appears.
Troubleshooting
| Symptom | Cause | Fix |
|---|---|---|
Operation denied by org policy, naming disableCustomMcpServerConnector | Google Cloud blocks custom MCP connectors by default | Allow custom MCP connectors for the data store's project. |
| Gemini Enterprise connects to Agent Authority, but Reload custom actions lists no tools, or tool calls return a permission error | No Agent Authority policy allows Gemini Enterprise's tool calls | Confirm a policy allows the agent |
| Failed to reload custom actions, with no error in Agent Authority | Google used a token issued before the connection was fully authorized | Click Verify Auth again, then retry Reload custom actions right away. |
| Gemini says the integration isn't connected, or answers from a web search | Gemini didn't choose the MCP data store | Name the MCP connection in the request. |
| The connection works at first, then stops | No refresh token, so the expired access token can't be renewed | In the data store's connection settings, check that Authorization URL Parameters is exactly &access_type=offline&prompt=consent and that Scopes includes offline_access. Correct either one, then click Verify Auth again. |
| No tool calls from a user appear in Agent Trail | That user hasn't clicked Authorize for the connector | Ask the user to open the connectors menu in the Gemini Enterprise web app and click Authorize next to the data store. Step 2 of Check it works shows where. |
Getting help
If you contact SecureAuth Support, include these details so Support can trace the request:
- The approximate time of the failure, with time zone
- The Client ID, never the client secret
- The names of the data store and the Gemini Enterprise app
- The exact error text, or a screenshot with credentials removed
- The step in this guide where the failure happened
Next steps
Gemini Enterprise's tool calls now run through Agent Authority from the data store you created, under each user's own sign-in. Allow all is fine for testing, but before users rely on the connection, allow only the tools Gemini Enterprise needs.
- Create a policy – control which tools Gemini Enterprise can call.
- Monitor activity – watch its tool calls appear in the dashboard and audit log.

