Sign in to continue learning
Build AI Apps with MCP Server: Working with Box Files
Intermediate
Topics
AI Frameworks
Agents
Data Processing
Document Processing
GenAI Applications
Prompt Engineering
Search and Retrieval
Task Automation
In this lesson, you'll learn how to make your solution more flexible and scalable using the Box MCP Server. All right, let's go. The solution you implemented works well as a proof of concept, but if you try to use it in a real business environment, you would run into some problems. The core issue is the custom built solution doesn't scale. First, in the last lesson, the files were previously downloaded to local. But having to download every invoice from Box before processing them is extremely slow and would not scale to processing thousands of files. Second, you built this script to handle specific invoice format. In reality, invoices come in various types, such as Word documents, PDFs, JPEGs, PNGs, and more. Your current code would require significant updates to handle that diversity. Also, as soon as the business needs to extract additional fields, you'll need to revisit and update the code. This makes the solution inflexible. And finally, processing files like this would need to be reviewed for best practices in data security due to storing files, often with sensitive data that would complicate this type of solution. This makes the solution impractical beyond a simple proof of concept. You'll need to give your current solution a boost in order to handle various file types seamlessly across the entire enterprise. This is where the Model Context Protocol, or MCP comes in. MCP will help you simplify the complexity of your application, making it more flexible and scalable. At its core, MCP is a standard that defines how context in terms of external tools and data resources is provided to AI applications. Right now, your solution was built custom with Gemini. You had to write custom one-off applications to process the invoices. If you wanted to extend the functionality and access other tools or data sources, you need to write more custom code. This is known as the M×N problem. For M applications and N external data sources or systems, you'll need to write M times N integrations, which quickly becomes unmanageable. By adding support for MCP, your application can access any external tool that also follows MCP's standard. Here's how it's done. The definition and execution of tools are offloaded to an MCP server. This means that inside your AI application, you no longer have to write custom code to connect your models to external systems. Instead, you use an MCP client within your application, which connects to the MCP server and sends tool invocation requests. The same server can be used with any other AI application. This simplifies everything into a much easier M+N setup. By using MCP, you can address many of the scaling and flexibility issues we identified before. Back to our use case. Let's look at a specific practical implementation that will help us improve the solution. Using MCP servers, such as Box's MCP server, we can outsource the integration work needed to almost any AI agent framework. We will use Box's MCP server for these examples, although this approach can be used with almost any MCP server and with other similar use cases. This is the fundamental power of MCP. We will be using a local MCP server that Box provides for this purpose. Essentially, it is a ready-made solution that handles all the complex interactions with the Box platform, including data extraction API, which is one of the most popular parts of Box AI API. Different vendors will offer different MCP servers, some of which you either access locally or remotely. Some other things Box MCP server provides. First, it allows AI applications to directly access, understand, and act on the content in Box. This means your application no longer needs to download and handle files. It can just ask the server to perform the actions on the content. The server is designed to handle large volume of documents from the start. It can execute operations on files and folders very efficiently. This addresses the speed and scale scaling issues in the initial solution. The server also includes built-in tools. It comes with capabilities for searching, querying data using AI, and extracting fields from documents. You don't need to worry about OCR code or prompting the model for specific fields, since these are now pre-built functions you can call. And finally, because it all happens within the Box platform, you have secure control over the files and the entire AI ecosystem. So here's the repo where you can find the files you can use to locally run the Box MCP server. You can see here the list of all the tools provided by the Box MCP server, which integrates the Box API to perform various operations such as file search, text extraction, AI-based querying, and data extraction. So let's go to the next lesson to rewrite the solution using Box's MCP server instead of the custom API code.
Build AI Apps with MCP Server: Working with Box Files