Run any SQL
SELECT for answers, and with Write access and your approval, DML and DDL for the teams that maintain marts here. Every write shows the statement first and ships on your yes.
BigQuery connector
BigQuery MCP Connector – Ask Your Warehouse in Plain English
Somewhere in your company is a BigQuery project holding the answers marketing keeps asking for: the GA4 export, the CRM sync, the revenue tables. Between marketing and those answers sits SQL, a data team's backlog, and a two-week turnaround for a one-line question.
The BigQuery connector removes the queue. It links the AI you already use, like Claude, ChatGPT, or Copilot, to your BigQuery projects through Mad Fish Elements. Your AI browses datasets, reads table schemas, writes the SQL your question implies, runs it, and returns the answer as a visual report in the chat. The warehouse your company already trusts becomes something marketing can simply ask.
See it
What Your AI Can See
Connect your projects, and your AI reads the warehouse the way an analyst would: list the datasets, list the tables in each, inspect any table's schema to learn what the columns actually mean. That exploration matters, because the difference between a right answer and a plausible one usually lives in the schema.
Then it queries. Any SQL the warehouse accepts: joins across the GA4 export and the CRM sync, window functions for cohort math, aggregations over billions of rows. Large results paginate cleanly, so big answers arrive without truncation surprises. You ask in plain language; your AI shows you the SQL it wrote alongside the answer, so your data team can verify the translation any time they like.
The cross-source power compounds here. The same Gateway reads your ad platforms live, so "join our ad spend against warehouse revenue by campaign" spans both worlds in one conversation.
Do it
What Your AI Can Do
The honest framing for most marketing teams: this is a read-and-analyze connector with writes available to the people whose job is writing. Read seats get the whole analytical surface and can change nothing, which is exactly how most analysts and marketers should start.
Explore safely
Dataset, table, and schema browsing that lets your AI ground its queries in what the tables really contain instead of guessing at column names.
Stream rows in
Insert rows into tables when a workflow feeds the warehouse, under the same approval flow.
Manage jobs
Long queries run as jobs your AI can track; the full v2 API is reachable through passthrough for load, copy, and extract jobs when a data workflow needs them.
A working session
The Two-Week Question, Answered Before Lunch
In practiceThe session that justifies this page: marketing wants customer lifetime value by acquisition channel, a question that has been in the data team's queue since last month. You ask it in plain language. Your AI lists the datasets, reads the schemas for the orders and attribution tables, writes the join, shows you the SQL, and runs it. The answer arrives as a chart: channel by channel, with the SQL underneath for your analyst to bless. What was a ticket became a conversation, and the data team reviews work instead of producing it from scratch.
The return
What You Get Back
The queue, gone
Plain-language questions against the warehouse, answered in minutes, SQL shown for verification.
Schema-grounded answers
Queries built from what tables actually contain, not from hopeful column guessing.
Spend beside revenue
Ad platforms live on one side, warehouse truth on the other, joined in a single conversation.
A data team doing data work
Analysts review and refine instead of fielding one-line questions, and marts get maintained under a real approval flow.
Use cases
One Conversation, Real Account Work
"What datasets and tables do we have in the marketing project?"BigQuery
The warehouse map, browsable in conversation.
"Show me the schema for the orders table."BigQuery
Columns and types, the grounding for every good query.
"Customer lifetime value by acquisition channel, last 12 months. Show the SQL."BigQuery
The two-week ticket as a visual report, with verifiable SQL.
"Join July ad spend from Google Ads against warehouse revenue by campaign."BigQuery · Google Ads
Spend-to-revenue truth across live platform and warehouse in one pass.
"Run the monthly cohort query and chart retention by signup month."BigQuery
The recurring analysis on demand or on schedule.
For in-house teams
Close the expectation gap
The warehouse stops being a place marketing asks about and becomes a place marketing asks. Read seats give the whole team safe analytical access; your data team keeps the write keys and reviews the SQL trail.
For marketing teamsFor agencies
Run the whole book
Clients with warehouses expect agencies to use them. Query client projects under per-client permission walls, ground your reporting in their revenue truth instead of platform-only numbers, and hand the client's data team every query you ran.
For agenciesReports worth sharing
From Live Account to Clear Story
Ask in chat and share the dashboard without building exports or taking screenshots.
Revenue Attribution Board
Warehouse revenue joined to live ad spend by campaign. Built for the meeting where marketing proves it.
Cohort and LTV Report
The analyses that never fit in a platform UI. Built for quarterly strategy.
Warehouse Freshness Check
Row counts and latest timestamps for the tables your reporting depends on. Built for trusting the numbers.
The charts do the talking: line for cohorts and trends, bar for channel comparison, table-first for the SQL-verified detail. And when your leaders or clients want these anytime, the Data Visualization add-on turns your reports into always-current dashboards they open from a link. Ask in chat. Share the dashboard.
A practical start
Your First Week
Day one: connect the project and look around; "List datasets and tables." No SQL yet, just the map. Day two: pick one table you care about. "Show me its schema." Ask one plain question against it; read the SQL your AI wrote. Day three: hand the answer to your analyst; let them verify the query. This step builds the trust everything else stands on. Day four: ask the question that has been in the queue. The cohort one, the LTV one, the channel one. Get the answer with the SQL attached; day five: schedule the recurring analysis. "First of each month, run the retention query and chart it." By Friday, marketing has asked the warehouse five real questions, the data team has verified the translations, and the ticket queue is shorter in both directions. That is the whole promise: not replacing the analysts, un-burying them.
Works best with
Build the Connected View
Pair BigQuery with Google Ads and Meta for spend-to-revenue joins that end attribution arguments. Add GA4 when the export lives here and the questions span raw events. The warehouse is where every other connector's data becomes one story.
Clear limits
What This Connector Will Not Do
It will not run costs blind. BigQuery bills by data scanned, and your AI writes bounded, schema-aware queries rather than reckless full scans; your project's own quotas and controls stay in force. It will not write without approval, and read seats cannot write at all. It will not invent data lineage; when a table's meaning is unclear, it says so and shows you what it checked. And it will not replace your data team; it clears their queue of the questions that never needed a ticket.
Governance
Governed Access Your Whole Org Can Live With
Access is set per person, not per account, so Read seats see everything and change nothing while Write seats create and edit with your approval flow in front of every change. Delete access goes only to people you trust to clean up, and many teams start read-only, check answers against the platform, then widen access as the audit trail earns it.
You choose the level of human review and can change it at any time, while every action records what changed, when, and who approved it. Per-user access means Elements never stores your BigQuery password, and your data is never harvested or resold.
Per-user accessRead / write / delete per seatFull audit trailApproval flow on every change
Setup
Setup Takes Minutes
- Start your trialCreate your Elements account and connect it to your AI.
- Link BigQuerySign in through the secure connection screen and choose what to connect. No code and no engineers.
- Set permissionsDecide who can Read, who can Write, and who can Delete. Start conservative and widen as trust builds.
- Ask your first question"How did we do last month?" is a complete instruction.
Plans, account limits, and the multi-client Agency plan are on the pricing page .
FAQ
BigQuery Connector Questions
Do my marketers need to know SQL?
No; they ask in plain language; the AI writes and shows the SQL. Analysts can read, verify, and refine every query, which builds trust faster than hiding the SQL would.
What about query costs?
Queries run in your project under your billing, quotas, and controls. The AI grounds queries in schemas and applies sensible bounds, and your data team can review the job history like any other usage.
Can it modify tables?
With Write permissions and approval, yes: DML, DDL, and row streaming for the teams that maintain marts. Most tenants keep marketing on Read seats and reserve writes for data staff, which is the intended shape.
Will my AI change things without asking me?
No. Your AI acts only with the permissions you set, and changes follow the review level you choose. Read-only seats cannot change anything, because the approval flow is the product working as designed.
Which AI assistants can our team use?
Your team can use Claude, ChatGPT, and Microsoft Copilot, while other AI tools can likely connect when they support MCP, the open standard for linking AI assistants to outside tools.
Is my data used to train anything?
No. Your data stays yours, so nothing is harvested and nothing is resold.
Focus on the Strategy, Delegate the Launch
The teams winning with AI are not handing their accounts to a black box; they keep the strategy and delegate the hands-on-keyboard work, while approvals and an audit trail make it safe to move fast. Connect the account you already have through the Gateway and ask the question you would normally build a report to answer.
Your first 14 days are free. Add your card to start, and cancel before the trial ends to pay nothing.