Upbuild
Written by Uptimal, the team that builds and operates Upbuild and runs marketing analytics for multinational clients across the globe. Everything below reflects how Upbuild actually behaves in production.
Your marketing data probably already lands in BigQuery: GA4’s native export, an ad-platform pipeline, a warehouse your data team owns. The gap has never been the storage, it is that answering “what did Search cost per conversion last quarter” means opening the console, remembering the table name, and writing SQL. Upbuild’s BigQuery family closes that gap: a small family of read-only tools that let Claude, ChatGPT, Copilot Studio or Gemini Enterprise reach the warehouse you already have.
What the assistant can actually call
bigquery_list_projects: the Google Cloud projects this sign-in can reach.bigquery_list_datasets: the datasets inside one project.bigquery_list_tables: the tables inside one dataset.bigquery_get_table: one table described, every column and its type, plus its size.bigquery_query: a read-only SQL query, rows returned into the chat.
That order matters. An assistant that has to guess a table name hallucinates one; an assistant that can list datasets, list tables and read a schema first writes SQL against columns it has actually seen.
Read-only by enforcement, not by convention
This is the one place in Upbuild where your assistant writes SQL, so the guard is not a promise in a system prompt. The statement is checked before it runs: only read queries are allowed (a statement that begins with SELECT or WITH), and anything that would insert, update, delete or change the structure of a table is refused outright. A prompt injection that tells the model to drop a table gets a refusal from the server, not a judgment call from the model.
Questions that work on day one
- “What tables are in my marketing dataset, and which one has last month’s spend?”
- “Describe the campaign performance table: what columns and types does it have?”
- “Top 20 campaigns by conversions last quarter from the warehouse, and how does that compare to what Google Ads reports today?”
The last one is the interesting one, and it is the reason a BigQuery MCP beats a SQL client: the same assistant holds your warehouse and your live Google Ads account in one conversation, so the reconciliation happens in the answer instead of in a spreadsheet afterwards.
Access is your access
Access is per person. The assistant reaches the Google Cloud projects your own Google account already reaches, nothing more, and a teammate connecting their own login gets their own reach. Nothing is copied into Upbuild: queries run in your own project, and the rows come back into the chat.
Ask your ad platforms directly, in Claude, ChatGPT, Copilot Studio or Gemini Enterprise.
Upbuild is an AI platform for marketing data: AI dashboards, insights, exports and alerts, with a read-only MCP server as part of its toolset that brings your accounts into Claude, ChatGPT, Copilot Studio and Gemini Enterprise. Every tool, live data. Free for 14 days on the connector and 7 days on the platform.
Get a demo Start free →Frequently asked questions
Does this replace the GA4 BigQuery export?
No, it reads it. If GA4’s native export already lands events in your project, these tools are how an assistant explores and queries that data conversationally.
Can the assistant write to BigQuery?
Not from the chat: bigquery_query only runs read queries. Scheduled exports into your data warehouse are part of AI Data Exports, set up by prompting Upbuild and run on a schedule, never by a chat message.
Whose BigQuery project does a query run in?
Yours. Queries execute in your own Google Cloud project under your own access, so billing and access controls stay where your data team put them.