Generative Manufacturing

Reimagining the factory floor through the lens of AI manufacturing

Comment

Generative Manufacturing

What it does

This project is an Model Context Protocol (MCP) server that provides a set of tools and prompts for controlling manufacturing. It allows Gemini to control a "micro-factory" to create and design and produce products. AI is also given the ability to monitor production of the products, allowing it to step in if the factory goes awry. The project includes a demo client that showcases the server's capabilities.

Inspiration

The inspiration for this project was to explore the interaction between AI and manufacturing. Specifically to see if AI could enhance the idea of "micro-manufacturing", the concept that we can achieve similar scale for manufacturing by taking a more distributed flexible approach rather than the monolithic massive operations approach that factories tend to be.

The benefit being that instead of shipping products across the world, we could save economically and environmentally by producing products that could easily be made locally at a Gemini augmented manufacturing facility.

I also wanted to explore the MCP standard and it's extension MCP Apps (aka MCP UI) as interesting new technologies on the market.

An AI Supervisor's Perspective

As an "AI Supervisor", I'm a firm believer in an AI-assisted future. I think that the way we interact with computers and machines is trending towards a new status quo, allowing us to take the role of a conductor or supervisor of systems. Generative Manufacturing captures the essence of what I believe the human-machine interface might evolve into.

Generative Manufacturing was built to work directly with my "micro-factory", my 3D printer. That being said the same principles that power Generative Manufacturing can extend to all kinds of machines, manufacturing domains and products.

What I learned

This project was perfect for learning new skills and concepts. I'm an avid user of MCP, and getting to try MCP Apps for the first time has opened the floodgates for my creative ideas. I dove deep into the MCP python SDK and continuing to understand how the MCP spec is built. I also got to learn more about Gemini 3, multi-model reasoning and how AI doesn't just have to be a chat bot.

I really got to see how AI fits into the manufacturing process. I was fortunate to really immerse myself in the world of 3D printing to learn how to not just use a 3D printer but how they actually work at a low level.

Accomplishments that we're proud of

I have a graveyard of unfinished projects and I'm sure I'm not the only person who operates that way. Proving to myself that I can get this project to the finish line, in a state that I can actually be proud of, brings a pride that keeps me motivated to keep doing what I do, not just in my coding, but in all parts of my life. This project will always be a remind of achieving daunting tasks as long as I put my mind to it.

How we built it

The project is built on a foundation of all kinds of technologies:

  • AI Studio and Gen AI SDK: Giving access to Gemini 3 AI greatness.
  • GCP: The project is deployed on GCP making it accessible to demo and use.
  • MCP Python SDK: This was the core library for building the MCP server.
  • MCP Apps (AKA MCP UI): Providing rich integrated UI for human interactions
  • TypeScript and the web: The MCP client is a web application written in TypeScript, providing a simple but powerful demo.
  • Openscad: Model generation using AI generated Openscad templates.
  • Prusaslicer CLI: Transformation of STLs to gcode that my 3D printer could understand using a Prusaslicer.

This project makes use of a number of Gemini 3 functionalities. Without the power of Gemini 3, this project would have failed spectacularly:

  • Thinking level: Changing the thinking level depending on the tool need. For example "quick print check" gets a low level of thinking to save bandwidth on high frequency calls. When more thought is needed is an incident occurs, high level of thinking is used like in "deep print check"

  • Media resolution: Same idea as Thinking Level where the high frequency quick tools can use low media resolution, where higher resolution images are passed when more detail is needed.

  • Function Calling: Allowing Gemini to call extra functions for 3D Printer information and stats if needed

  • Multimodal functionality: Whether it's looking at images to monitor a print or it's generation of Openscad templates for 3D model generation, Gemini 3 multimodal was really put to the test.

Challenges I ran into

MCP Apps is a very new extension for the MCP spec. This means there is almost no support for it in most MCP clients. I spent more time than I would have liked trying to find something that would actually show the UI correctly for the video (hence why I was stuck with Claude Desktop as the chat interface in the video). It was even harder to find something that I could host a demo with. My initial plans were to create my own MCP client/host but it would have taken me more time than I had to the deadline.

Another significant challenge was the 3d model generation and slicing. A significant chunk of time was spent reverse engineering undocumented functionality to achieve a truly end to end AI driven workflow that I showcased.

What's next for Generative Manufacturing

This is the very beginning of where Generative Manufacturing can go. This is one of the few ideas I've had that could really show how AI integrates seamlessly with the real world. Here are a few ideas that would take this concept to the next level:

  • Control of entire 3d print farms
  • Control of other machines like robot arms and CNC Routers
  • Deployment in a live factory line
  • Weeks to months long automatic production monitoring (I only had time to test minute timelines)
  • Predictive and proactive maintenance monitoring

Generative Manufacturing can really manufacture the future!

Built With

Share this project:

Updates