MCP Toolbox for Databases in Action

Key Takeaways

[Music] Maybe you'd like to develop agents to understand natural language requests and even have them dynamically select the right tools based on your data. Connecting these intelligent agents to databases can often feel like navigating a maze. You're going to have to break down silos, ensuring robust security, all that feels challenging. And this is where MCP toolbox for databases comes in. It's an open-source MCP server designed specifically to make this connection easier, faster, and more secure. MCP or model context protocol acts as the standardized way for AI agents to communicate with tools, much like a universal adapter. So why should you consider using the toolbox? First of all, it simplifies your development. You can integrate tools in your AI agent with very little code. Second, it delivers better performance because Toolbox incorporates best practices like connection pooling right out of the box. Third, security is enhanced with integrated authentication methods. You also gain endtoend observability thanks to out-of-the-box metrics and tracing with built-in open telemetry support. And finally, MCP toolbox provides a centralized location for your tools, allowing you to share them between agents and applications. Let's look at the architecture. MCP toolbox acts as a control plane sitting between your application's orchestration framework. And this could be something like Google's ADK agent development kit which acts as the MCP client and the MCP toolbox which is your server exposing those defined database capabilities. To see this in action, let's explore an example, the time series forecasting agent. You can find this project in the GitHub ADK samples repository. It shows you how you can build an intelligent agent using Bitquery's AI forecast function which uses a state-of-the-art times FM model. This agent can understand natural language requests for forecast and dynamically select the right tools. So imagine you want to predict future values based on historical data. In this case, we could ask the agent forecast liquor sales for the next week in Iowa. The agent potentially powered by a large language model like Gemini first needs to understand this natural language request and then it figures out how to get that forecast and this is where the tools come in. The forecasting agent is configured to load its tools from the toolbox which acts as the MCP server. These tools are defined in a config file called tools.yml. Each tool has a description that the agent uses to understand its purpose. It will also list any parameters needed like horizon in our case and contain the actual SQL statement that interacts with the database. The really powerful aspect here is the dynamic tool system. You can add or update forecasting capabilities simply by updating that tools.yml file and then restarting the MCP server. So there's no need to change the agents core Java code, recompile, any of that. This makes the entire system highly extensible and maintainable. We can interact with this forecasting agent using the ADK's forecasting UI. You'd first type in your forecasting request. The agent with the help of the MCP toolbox understands this, selects the correct tools, determines parameters like the horizon, and then instructs the toolbox to execute the query. So the agent not just presents raw forecast data but it provides qualitative analysis and insights and this gives us an interactive way to experiment and see the agents reasoning and results in real time. So this agentic design is particularly effective because it combines the broad contextual understanding of a large language model with the specialized precision of a model like the times FM model for forecasting. MCP toolbox for databases currently supports a broad number of databases from Google Cloud and even includes contributions for thirdparty databases like Neo forj and Dgraph for integrating these tools in your applications. You also have a client SDK and integrations for popular frameworks of lingraph in llama index. To sum it all up, MCP Toolbox for databases is an open- source MCP server that drastically simplifies how your AI agents can utilize enterprise data. If you're building agentic applications that need to interact with databases, this is a tool you'll want to check out. Visit the MCP toolbox for databases on GitHub. Explore this documentation and see how it can accelerate your AI development today.