What Was Announced
Google Cloud released a Java SDK for the MCP Toolbox for Databases in the first half of March 2026. The MCP Toolbox is Google's Model Context Protocol server that provides AI agents with a standardised interface to connect to databases — and until this release, the tooling was primarily Python-centric. The Java SDK changes that, making it a first-class option for Spring Boot and LangChain4j developers who want to give their AI agents database access without leaving the Java ecosystem.
Key Features and Changes
What Is MCP Toolbox for Databases?
The MCP Toolbox is a server that sits between an AI agent (such as Claude Code, Gemini, or a custom LangChain4j agent) and one or more databases, translating natural-language or structured AI requests into safe, connection-pooled database queries. It supports tens of data sources including Cloud SQL (PostgreSQL and MySQL), Cloud Spanner, AlloyDB, BigQuery, and Firestore.
Java SDK — Spring Boot Integration
The Java SDK provides a Spring Boot auto-configuration that wires the MCP Toolbox client into the Spring context with minimal setup. You declare your data sources in a configuration file, add the Spring Boot starter dependency, and the toolbox is available as a bean that your AI agent layer can call.
<dependency>
<groupId>com.google.cloud</groupId>
<artifactId>spring-boot-starter-mcp-toolbox</artifactId>
<version>1.0.0</version>
</dependency>
LangChain4j Integration
For teams using LangChain4j as their Java AI orchestration layer, the SDK provides native tool definitions that LangChain4j can discover and register automatically. This means your @Tool-annotated AI agent methods can delegate to the MCP Toolbox for any database query without manual JDBC or JPA boilerplate.
High-Concurrency Design
The SDK is specifically optimised for the high-concurrency patterns typical of AI agent workloads — where many parallel requests may query the same database simultaneously. The connection pooling, timeout, and retry configuration is tuned for agent use cases rather than traditional request-per-thread models.
Firestore Regional Endpoints GA
In the same release period, Google also announced GA of Regional and Multi-Regional endpoints for the Firestore API. Applications can now ensure their Firestore requests are transmitted, stored, and processed entirely within a specific geographic region or multi-region, which matters for data residency compliance requirements.
Why It Matters for Developers
Before this release, Java developers who wanted to give AI agents database access had two options: write their own MCP server in Python and call it from Java, or implement ad hoc tool methods that bypassed MCP entirely. Neither is ideal for production use.
The Spring Boot starter removes that friction. Java teams building AI-augmented applications — chatbots that can query customer data, assistants that can analyse job queues, or agents that generate reports from BigQuery — can now use the same Spring Boot patterns they already know, with Google Cloud managing the database connectivity and security layer through MCP.
Source
Read the original article — Google Cloud Blog
