Simplify AI Agent Access to Enterprise Data with the MCP Toolbox for Databases
Google Cloud revealed amazing ways for businesses to create multi-agent ecosystems using Vertex AI and Google Cloud Databases at Google Cloud Next 25. These include improved methods for agents to interact with one another through the use of Agent2Agent Protocol and Model Context Protocol (MCP). Google is making it simple for MCP Toolbox for Databases (previously Gen AI Toolbox for Databases) to access your corporate data in databases in light of the increased interest in MCP for developers. This is an additional advancement in offering standardized and safe methods for experimenting with agentic applications.
Previously known as Gen AI Toolbox for Databases, MCP Toolbox for Databases
An open-source MCP (Model Context Protocol) server called MCP Toolbox for Databases (Toolbox) enables developers to safely and simply link new AI agents to corporate data. Anthropic developed MCP, a new open standard that replaces fragmented connections that call for custom integrations by establishing a standardised interface to link AI systems with data sources.
At the moment, Toolbox may be used to create tools for a wide range of databases, including self-managed MySQL and PostgreSQL, Spanner, Cloud SQL for PostgreSQL, Cloud SQL for MySQL, and AlloyDB for PostgreSQL (including AlloyDB Omni). It incorporates contributions from third-party databases like Neo4j and Dgraph as it is entirely open-source. Toolbox provides end-to-end observability with OpenTelemetry integration, improved security with OAuth2 and OIDC, and easier development with less boilerplate code. By managing the intricacies like connection pooling, authentication, and more, this helps you to create tools more quickly, easily, and securely.
The extra scaffolding needed to create production-quality database tools and make them available to every client in the expanding MCP ecosystem is provided by Toolbox, an MCP server. This compatibility streamlines development and improves interoperability by enabling developers creating agentic apps to use Toolbox and safely query a variety of databases using a single, standardised protocol.
Agent Development Kit (ADK) is supported by MCP Toolbox for Databases
The Agent Development Kit (ADK), an open-source framework that makes it easier to create complex multi-agent systems while preserving fine-grained control over agent behaviour, was then introduced. You can create an AI agent using ADK in less than 100 lines of user-friendly code. ADK allows you to:
- Use orchestration controls and deterministic guardrails to influence your agents’ reasoning, thinking, and teamwork.
- With only a few lines of code, you can use ADK’s exclusive bidirectional audio and video streaming features to have human-like interactions with your agents.
- Select the deployment or model that best suits your requirements. Whether it’s your preferred top-tier model, deployment target, or integration with remote agents built on other frameworks, ADK works with your stack of choice. Additionally, ADK supports the Model Context Protocol (MCP), which allows your data sources and AI agents to communicate securely in both directions.
- Use the direct interface with Vertex AI Agent Engine to deploy to production. The usual overhead involved in putting agents into production is removed by this dependable and transparent route from development to enterprise-grade deployment.
MCP Toolbox for Databases
Including support for LangGraph
With the help of checkpointers, LangGraph provides you with crucial built-in support for the persistence layer. This aids in the development of robust, stateful agents that can consistently do lengthy tasks or pick up where they left off.
Google Cloud provides specialised integration libraries to take use of robust managed databases for storing this state. The following options are available to developers:
- Use the AlloyDBSaver class from the langchain-google-alloydb-pg-python library to access the highly scalable AlloyDB for PostgreSQL, or choose
- Cloud SQL for PostgreSQL using the langchain-google-cloud-sql-pg-python library’s PostgresSaver checkpointer implementation.
Supported by the performance and management of Google Cloud’s PostgreSQL products, both provide strong methods to smoothly store and load agent execution states, enabling processes to be interrupted, resumed, and audited with reliability.
A checkpointer records a checkpoint of the graph state at each super-step when you use it to assemble a graph. These checkpoints are stored to a thread that is accessible following the execution of the graph. Threads provide a number of potent features, including as fault-tolerance, memory, time travel, and human-in-the-loop, by providing access to the graph’s state after execution.
