10. Create a MCP server

I have a friend who works (like me) as a software engineer

I call him a few times a week, and we talk about things going on in our life, tech, work - a whole host of things

He told me recently (because he's started a new job) about some cool AI stuff he's been doing

I was intrigued, and wanted to try and do two things with my next project

  1. I wanted to build something with AI
  2. I wanted to build something for AI

When I say "for AI" I mean, I wanted to try and make a Model Context Protocol (MCP) server

We've "solved" essentially how to nicely prompt machines with language models, but one problem we currently face is the data we use to train these models is so large they aren't specialists

These language models can act like a therapist, a friend, a devil's advocate - whatever you want

An MCP server is a protocol for allowing AIs to search the web, query data, run terminal commands, do automated tasks etc.

It's augmenting AI to be much, much more powerful - and I wanted to try and make my AI much more specific, over a small area of domain knowledge, and so here began project 10!

What data am I gonna use? 🤔

I found a database of the Cassini-Huygens space-research mission by NASA

This was a 13 year mission, between 2004-2017, and was created to study the planet Saturn and its system, including its rings and natural satellites

The probe orbited Saturn 294 times, and the database of logs is free and has approx 61,874 activities logged

An image of a SQLite database with a large amount of rows, and logs shown

This looks like a good start!

Can an AI create the MCP server?

I wanted to try and ask Claude to break this server down into 5 iterations, and do each section at a time

I also asked it to create audit logs (of what I prompted) and what it did in my docs folder, which it did successfully, you can view here

I can now just prompt, and let Claude do most of the work here

Okay, iteration 1 through 5 is now complete - let's test this with our client

Some examples of client prompts

A prompt from AI explaining the building of an MCP server, was a success

An AI prompt asking how many teams were involved in the Cassini-Huygens mission, and the answer being 15

An AI prompt asking what the Magnetometer team did

An AI prompt asking how many flights there were, and the answer being 61,874

An AI prompt asking what the most common type of flight was, and the answer being SPASS Rider, SPASS meaning Science Planning and Sequencing Software.

An AI prompt asking what the most common types of flights were, and the answer 1. SPASS Rider, 2. Non-SPASS, 3. Prime, 4. SPASS Note, 5. (Empty/Null) and 6. spass_type

An AI prompt asking how many Prime flights there were and the answer being 13,600

An AI prompt asking if the AI could answer as clearly without the MCP server, and the answer being it could not

This has been a pretty interesting learning exercise for me to learn more about AI

So far, I'm a little pessimistic and largely agree with Linus Torvalds who said AI coding is fine, as long as it's not being used in production - but I'm intrigued to see how development cycles change over the coming years

We've officially reached 1/3 of the way through 30 projects!

Onwards dear knight, to project 11 ✔️

←

Jump to project

12345678910