Reads
Answer straight away. Nothing changes.
- Table metadata and schemas
- Constructed SELECT dry runs
- Status and bounded results for this account's admitted jobs
Analytics · MCP server
BigQuery MCP server
Review query costs before execution.
MegaMCP is a hosted BigQuery MCP server that lets Claude, ChatGPT and Cursor read dataset tables, schemas and query cost estimates and propose query runs with an explicit byte limit that you approve.
Rolling outBuilt, and switching on as approvals land.What this meansRolling out means the code is built and the platform is still approving MegaMCP's app: Google, Meta, Microsoft and the one-click ad-platform sign-ins switch on as each approval lands.
- Tools
- 5 reads, 1 write
- Works with
- Claude, ChatGPT, Cursor, Claude Code and VS Code
- Price
- Free to read, changes from $19/mo
BigQuery5 reads · 1 changeMegaMCPApproval gate
- Claude
- ChatGPT
- Cursor
Live demo
Watch it work on BigQuery
Ask in plain English. Your AI reads BigQuery with real MegaMCP tools, then hands you a plan. Nothing moves until a person clicks Approve, and supported changes can be undone.
Tools in this example
bigquery_dry_run_selectReads live databigquery_execute_selectProposes a change plan
- You
Total cost by campaign for September from the ads table. Keep it under 2 GB.
- Claude
bigquery_dry_run_selectDonescans 1.2 GBbigquery_execute_selectWaiting for approvalplan readyA dry run says this query scans 1.2 GB. Here's the run, with a 2 GB byte limit.
Run a queryBigQuery · marketing.ads_daily3 changes- Add: SELECT campaign_id, SUM(cost) … GROUP BY campaign_id
- Add: Maximum bytes billed2 GB
- Flagged: Running a query can't be undone
- Claude
Done. 38 rows, 1.2 GB billed.
What your AI can do
BigQuery MCP tools
Inspect an enabled dataset, construct a bounded SELECT, and review its dry-run byte estimate before approving execution. This list comes straight from the running product: 5 read tools and 1 write tool.
5 tools
Changes you approve
Each one returns a plan with a diff. Nothing moves until a person approves.
- Human-approved constructed SELECT execution with explicit cost consent
BigQuery guardrail
Reads never execute paid queries. Execution is irreversible and byte-limited, not dollar-limited. Raw SQL, joins, views, external tables, DDL and DML are unavailable.
Plus complete diff previews, human approvals and the flight recorder on every connection. Live changes require a paid plan.
All 6 BigQuery tools5 reads, 1 change · straight from the running product
| What it does | Tool id | Type |
|---|---|---|
| List client dataset tables | bigquery_list_tables | Read |
| Inspect allowed native table | bigquery_get_table_schema | Read |
| Estimate bounded SELECT cost | bigquery_dry_run_select | Read |
| Propose bounded query execution | bigquery_execute_select | Change |
| Reconcile approved query job | bigquery_get_query_job | Read |
| Read approved query results | bigquery_get_query_results | Read |
Example prompts
BigQuery MCP prompts
Paste any of these into Claude, ChatGPT or Cursor once BigQuery is connected.
“List native tables in this dataset.”
“Dry-run these selected columns and filters.”
“Prepare this query with an explicit maximum bytes billed.”
“Show the schema of the events table.”
“How many bytes would this query scan?”
“Which columns hold campaign IDs?”
“Run the approved query with a 1 GB maximum bytes billed.”
“Read the results of the query I approved.”
Setup
BigQuery MCP setup in 3 steps
About two minutes for Claude, ChatGPT or Cursor. Nothing to install.
Connect BigQuery
Start a free workspace and open Connections. Requires Google BigQuery consent, enabled API and dataset/table access plus query-job permission. The billing project is the chosen dataset project.
Add MegaMCP to your AI app
Every app uses the same server URL.
MCP server URLhttps://megamcp.com/api/mcpAsk in plain English
Try “List native tables in this dataset.” Reports answer straight away. Anything that changes BigQuery comes back as a plan for you to approve.
Safety
Hand AI the keys. Keep the brakes.
The same loop guards every MegaMCP connection: ask, see the diff, approve, undo.
Approval before every change
BigQuery write tools never touch BigQuery directly. They return a plan, and a person approves it in the dashboard, or in chat if an admin allows it.
Approval requestedBigQuery · 1 plan waitingDashboard · just now
A diff first
Every plan lists each change with its before and after values, so you see exactly what will move before it does.
Plan previewBigQuery1 change
- Change: Value
BeforeAfter - Check passed: Checked again right before it applies
Undo for supported changes
Supported changes reverse through a separately reviewed undo plan. Anything that can't be undone is marked before you approve it.
Undo plan readyRestores the previous valuesReviewed like any other change
Receipts for everything
Platform checks stop stale edits, and the flight recorder logs every tool call, plan and approval, with who did it and from which AI app.
| Time | Event |
|---|---|
| 09:14 | bigquery_list_tables |
| 09:15 | Plan created by Claude |
| 09:16 | Approved by Ana |
FAQ
BigQuery MCP server questions
Is there an official BigQuery MCP server?
Yes. As of October 2026, Google runs a managed remote MCP server for BigQuery. MegaMCP takes a cost-first approach for marketers: it dry-runs a bounded SELECT, shows the bytes it would scan, and runs it only after a person approves an explicit byte limit.
What is a BigQuery MCP server?
An MCP (Model Context Protocol) server lets AI assistants such as Claude, ChatGPT and Cursor use BigQuery through tools. MegaMCP's hosted BigQuery MCP server gives your assistant 5 read tools and 1 write tool for dataset tables, schemas and query cost estimates, and every change waits for a human approval.
Does the BigQuery MCP server work with ChatGPT?
Yes. MegaMCP is a remote MCP server with OAuth sign-in, so one URL works in ChatGPT (as an app, with developer mode on), Claude (as a custom connector), Cursor, Claude Code and VS Code. Your AI app's own plan decides whether it can add custom MCP servers.
Can it make changes to my BigQuery account?
Yes, with your approval. It can propose query runs with an explicit byte limit changes. Each one comes back as a plan with a line-by-line diff and waits for a human before anything moves. Reads never execute paid queries. Execution is irreversible and byte-limited, not dollar-limited. Raw SQL, joins, views, external tables, DDL and DML are unavailable.
How do I connect BigQuery to Claude or ChatGPT?
Requires Google BigQuery consent, enabled API and dataset/table access plus query-job permission. The billing project is the chosen dataset project. Public OAuth verification and live acceptance remain pending.
What does the BigQuery MCP server cost?
Reading BigQuery is free on MegaMCP's Free plan (100 reads a day, no card). Approved changes are included in every paid plan, from $19/month after a 5-day free trial, with unlimited runs and no credits. MegaMCP adds no markup: any subscription or API usage with the vendor stays between you and them.
Is my BigQuery data safe?
Your BigQuery credentials are encrypted at rest (AES-256-GCM) and bound to their connection, and MegaMCP never trains AI models on your data. Your AI only sees the accounts you switch on, members limited to certain clients only see those clients, and every tool call lands in the flight recorder.
Keep exploring
More MCP servers
More analytics and seo
Google Analytics MCPRolling outAsk GA4 questions in plain English.
Google Tag Manager MCPRolling outFix tracking without spelunking GTM.
Google Search Console MCPRolling outYour SEO data, finally conversational.
Semrush MCPLiveResearch a domain with bounded API credits.Ahrefs MCPLiveInspect search and backlinks with a clear allowance.DataForSEO MCPLiveRead your existing research receipts.
PageSpeed Insights MCPLiveRun Lighthouse checks on your public site.Bing Webmaster Tools MCPLiveRead Bing performance for enabled client sites.Popular integrations
Google Ads MCPRolling outRun Google Ads from one sentence.
Meta Ads MCPRolling outMeta Ads, minus the Ads Manager maze.
HubSpot MCPLiveClose the loop between ads and pipeline.
Shopify MCPLiveAsk your store how it's doing.
WordPress MCPLiveFix your WordPress site by asking.Learn more
5 days. On the house.
Put Mega to work. Your ad budget stays yours.
No credit card to start. Every workspace comes with a sandbox ad account, so you can watch the whole ask, diff, approve, undo loop before you connect anything real.