A new course, thanks to Andrew Ng and his partnership with Box, that shows how you can leverage MCP servers to offload otherwise laborious and custom-made work.

And this is done with a real world example of using Box's MCP server to extract information from pdf invoices residing in a Box folder. This is an elaborate task since you have to parse and work with unstructured data which is the "worst" kind of all. The story doesn't end there though. You'll also make and coordinate Agents using Google's ADK and the A2A protocol. 

But before getting into the advanced parts, and in order to showcase the usefulness of the MCP-based solution, Ben Kus CTO at Box who serves as the instructor of the course, first builds the application by directly interfacing with the Gemini LLM from Python code. The truth is that while the solution doesn't look complicated. it won't scale in real world scenarios.

However, in Part Two he shows how an MCP server tackles the MxN problem of scaling, that is that without MCP when wanting to use another tool, then you have to write new code and custom connectors to do that. With the MCP server in place you just use a MCP client as the intermediary which forwards requests to the Server and which subsequently decides which tools to use to perform the task. And any app that adheres to the MCP protocol can hook up to it seamlessly.

This is backed by a replaying the previous non-MCP example, but this time by using Box's MCP server, where using the minimum amount code you offload all the work to the server; extracting information and returning it back as structured fields.

Finally we get into the Agentic AI, where you restructure your previous MCP application to use a multiagent architecture where each agent is carrying out a specific task. This in the non AI world is analogous to the Microservices Architecture which sponsors composability, independence, autonomy, scalability, easier observability and debugging. Of course as Microservice APIs do, independent Agents need to be coordinated. There's where the A2A protocol and Google's ADK step in.

In conclusion, I'll tell you what the real magic with this course is. Even if you never plan to use Box's MCP server, in just 36 minutes you'll get the gist of what MCP and Agentic AI are all about and where they differ; concepts important in this age of AI applications.

More Information

Build AI Apps with MCP Servers: Working with Box Files 

Related Articles

Build Apps with Windsurf's AI Coding Agents - The Course 

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info