Model Context Protocol is changing how we interact with our home computers, not just cloud services.

Let me explain: Right now Model Context Protocol supports Streaming HTTP, which is primarily how you would connect to MCP servers in the cloud but they also support booting up processes running locally on your computer and communicating with that via STDIO. Great, what does this mean?
This means your LLM agent running locally can not only boot up tools that you need to run locally such as your video editor, or 3d modeling software but it can also interact with those tools in a native way as well.
I am sure we are all hearing about MicroSoft giving access for LLM agents to view and click stuff on your desktop. Scary right? Especially because the LLMs agents could click on anything, or mistake some button that a human could easily recognize as “DELETE ALL” for “SAVE”. This is clumsy and prone to error.
With MCP(or a similar protocol) you could give the LLM agent direct access to only the programs you wanted it to have access to. The burden of guessing as to what was on the screen and what to click on would be removed.
Lets dig a little deeper: You tell the LLM “Boot up my video editor” so it spawns a process that is using MCP’s STDIO feature to communicate. The video editing software then boots up and once ready sends the schema of tool
commands available to your LLM Agent.
You then tell the agent “Import the footage from my latest shoot and organize it like my other projects”. The LLM looks at the available tool
commands and decides to first import
then to transcribe
each clip so it can finally read through all the transcripts and decide how to organize them based on your previous projects. All while you sit back and enjoy a cup of coffee.
If you would have asked me even a few months ago I would have thought this to be science fiction but now that I have gotten in deep I think it's coming faster than we could have imagined.
If you want to jam with me about where this all is going please sign up for my Workshop on how to run AI/ML on AWS this Friday at 1PM Central and we can chat about it then.
Don't give it access to the nukes.