BigQuery

Diskret connects to Google’s managed BigQuery MCP server (https://bigquery.googleapis.com/mcp). It is switched on together with the BigQuery API, so there is nothing to deploy. Each user signs in with their own Google account, and BigQuery’s IAM decides which projects and datasets they can see.

Time: about 15 minutes. Who: someone who can grant IAM roles in the Google Cloud project that holds the data, and a Diskret admin.

What Diskret can access

Scope	Why	Required
https://www.googleapis.com/auth/bigquery	List datasets and tables, read their metadata and run queries	Yes

This scope reads “view and manage your data in BigQuery”. It is the only scope the server offers, so IAM roles are what keep access read-only (step 2): with BigQuery Data Viewer and no editor role, a user cannot change tables, whatever the AI asks for.

Tools

Tool	What it does	In Diskret
list_dataset_ids	List datasets	Open
list_table_ids	List tables in a dataset	Open
get_dataset_info	Read a dataset’s metadata	Open
get_table_info	Read a table’s metadata and schema	Open
execute_sql_readonly	Run a query; DML, DDL and Python UDFs are refused	Open
execute_sql	Run any SQL, including statements that change data	Keep closed

How personal data is protected

Query results are free-form: the AI chooses the columns, aliases and joins, so a result has no fixed field paths for Diskret’s field rules to match. Diskret therefore scans every value in every result and pseudonymises what it finds. That works, but it is the weakest setup Diskret offers, so keep the data the AI can reach small:

• Grant BigQuery Data Viewer on the datasets users need, not on the whole project.
• Prefer datasets without personal data: aggregated reporting tables, or views that leave out names, email addresses and phone numbers.
• Keep raw exports (CRM leads, form submissions, user-level analytics) in datasets the AI cannot read.

Before you start

• The BigQuery API is enabled in the project (it is by default in new projects). Enabling it needs the Owner or Service Usage Admin role.
• You have an OAuth client in Google Cloud. If you already set up Google Workspace, reuse that client.
• Turn off any direct BigQuery connector in Claude once Diskret is set up. See Recommended Claude settings.

1. Enable the API

Open the Google Cloud console, select the project that holds the data, go to APIs & Services → Library and check that the BigQuery API (bigquery.googleapis.com) is enabled. The MCP server comes with it.

2. Grant IAM roles

Grant these to each user, or better to a Google group the users belong to:

Role	Where	Why
MCP Tool User (roles/mcp.toolUser)	Project	Call MCP tools
BigQuery Job User (roles/bigquery.jobUser)	Project	Run queries
BigQuery Data Viewer (roles/bigquery.dataViewer)	Each dataset users need	Read the data

Do not grant BigQuery Data Editor, BigQuery Admin or Owner to users who query through AI.

Cost: queries are billed to the project named in the query’s project_id, at your normal BigQuery prices. A query runs for at most three minutes and returns at most 3,000 rows.

3. Set up the OAuth client

If you reuse the Google Workspace client, only add the scope:

1. Go to Google Auth Platform → Data Access, click Add or Remove Scopes, paste https://www.googleapis.com/auth/bigquery under Manually add scopes, click Add to Table, then Update, then Save.

If you need a new client, follow steps 2 and 3 of the Google Workspace guide: an Internal consent screen, then a Web application client with Diskret’s Redirect URI under Authorized redirect URIs. Copy the Client ID and Client secret.

Caution

Google’s BigQuery MCP documentation says custom redirect URIs are not supported, without saying more. We read that as custom URL schemes (such as myapp://), not an ordinary https redirect URI like Diskret’s, but we have not verified it yet.

4. Add the connector in Diskret

1. In Diskret, go to Connectors → New connector. Type: MCP server.
2. Choose the Owner unit, name it BigQuery, slug bigquery, and set the URL to https://bigquery.googleapis.com/mcp.
3. Under Authentication, choose Sign in with OAuth, and under Whose credential choose Personal (each user connects their own).
4. Paste the Client ID and Client secret. Under Scope, enter https://www.googleapis.com/auth/bigquery.
5. Save, then click Check sign-in. Diskret shows the authorization server it found and the scope it will ask for.
6. Under Tools and access, open the read tools and execute_sql_readonly for the units that need them. Keep execute_sql closed.
7. Optional: to limit which Google Cloud projects each unit can query, set Scope argument to project_id and allow the project ids per unit under Scope. Nothing is allowed until a scope is set.

Caution

execute_sql_readonly and execute_sql take a project_id argument. Check after Refresh tools whether the list and metadata tools take it as well, before you rely on the scope argument for all tools.

Each user then goes to My connections, clicks Connect and signs in to Google. Queries run with that user’s own IAM roles and appear in BigQuery’s job history and audit logs under their name.

Maintenance

• Access: add or remove users in the Google group from step 2. Removing BigQuery Data Viewer on a dataset takes effect on the next query.
• Revoke access: remove the connector in Diskret, or delete the OAuth client in Google Cloud (this also cuts off every other connector that uses it). Users can revoke their own access with Revoke under My connections.

Troubleshooting

Error	Likely cause	Fix
Permission denied on mcp.tools.call	The user lacks MCP Tool User	Grant roles/mcp.toolUser on the project
Permission denied on bigquery.jobs.create	The user lacks BigQuery Job User	Grant roles/bigquery.jobUser on the billing project
Access denied on a table	The user lacks BigQuery Data Viewer on that dataset	Grant roles/bigquery.dataViewer on the dataset, if the user should see it
invalid_scope	The scope is missing on the consent screen	Add it under Data Access
Query cancelled	It ran longer than three minutes	Narrow the query, or query a smaller or pre-aggregated table
Result seems cut off	Results stop at 3,000 rows	Aggregate in SQL instead of returning raw rows
Query on a Drive-backed table fails	External tables on Google Drive are not supported	Query a native table instead