BigQuery Read-Only MCP Server
Open-source MCP server connecting AI agents to enterprise BigQuery data — with hard table allowlists, per-query scan ceilings, and built-in rate limiting.
I build open-source MarTech pipelines on Google Cloud.
BigQuery, BQML, GA4, and the Model Context Protocol — connected, hardened, and shipped to production. Follow me on GitHub for more to come.
Projects
GA4 Ecommerce BQML Purchase Propensity
Self-hosted BigQuery ML pipeline that predicts purchase propensity from GA4 events and pushes the result back to GA4 as a user property via the Measurement Protocol. Built for Google Ads remarketing.
GA4 BigQuery ML Sales Forecast
A pure-SQL pipeline that turns the GA4 BigQuery export into a 30-day daily revenue forecast — with custom holidays, automatic seasonality, ad-spend regressors, and seasonal sub-forecasts for upstream funnel metrics. Output is a single view you connect directly to Data Studio.
GA4 Anomaly Detector
Statistical anomaly detection over GA4 BigQuery exports, narrated by an LLM. The pipeline does the math in code (STL, PELT, JS-divergence); the LLM only writes the prose. Includes a CLI and an MCP server for Claude Desktop, Cursor, and Gemini CLI.
Recent posts
-
Forecasting GA4 revenue in pure SQL with BigQuery ML — and the model-quality checks you can't skip
— May 18, 2026
A pre-release SQL-only pipeline produces a 30-day GA4 revenue forecast with ad-spend regressors and seasonal sub-forecasts — no Python, no Vertex, no Cloud Functions. The archit...
-
When +35% is really −60%: a GA4 anomaly detector where the LLM only writes the report
— May 18, 2026
I've been using MCP servers for about a year, like the rest of the tech industry. Plenty of impressive moments. Also plenty of cases where the LLM confidently does a calculation...
-
Self-hosted vs Google's official BigQuery MCP server: a security and cost comparison
— May 18, 2026
Google's official BigQuery MCP server is the right default for trusted internal analysts. For agents handling untrusted input, a self-hosted server with hard table allowlists an...
-
How to build a GA4 purchase propensity model with BigQuery ML (without Vertex AI)
— May 15, 2026
GA4's built-in predictive audiences need ≥1,000 returning purchasers in the last 28 days. If you don't have that traffic, you can roll your own with BigQuery ML — no Vertex AI, ...
-
Stopping the $2,000 AI query: how to cap BigQuery scan cost from an MCP server
— May 11, 2026
Every AI agent connected to BigQuery is one bad query away from a four-figure invoice. There are three layers of defense; only one of them is bulletproof. Here's where to put each.
What I write about
BigQuery · BigQuery ML · Google Analytics 4 · Model Context Protocol · Cloud Run · Google Ads · Measurement Protocol · Marketing data engineering.