Member-only story
MCP with BigQuery: A Revisit
With the recent release of fully managed, remote Model Context Protocol (MCP) servers for Google services, AI agents can now securely and directly interact with BigQuery using a standard, tool-based interface — no bespoke integrations required. In this post, we’ll walk through how the BigQuery MCP server works, why it matters for data-driven agents, and how to build a working analytics agent using Google’s Agent Development Kit (ADK) and Gemini.
Why MCP Changes the Game for Data Agents
Most “data-aware” LLM applications today rely on indirect access patterns:
- Prompted SQL generation
- Custom REST endpoints over warehouses
- Middleware layers that sanitize, validate, and execute queries
These approaches work — but they’re brittle, hard to standardize, and expensive to maintain.
Model Context Protocol (MCP) introduces a different abstraction. Instead of forcing agents to hallucinate APIs or SQL dialects, MCP exposes enterprise systems as typed tools with a well-defined contract. The agent doesn’t “guess” how to query BigQuery — it calls a BigQuery tool.
With remote MCP servers, those tools live on the service’s own infrastructure and are accessed via HTTP. This gives you:
