Introduction

Overview of MCPs

I wanted to talk a little bit or I wanted to demo MCPs or Model Context Protocols today a little bit.

How many of you know what MCP is, Model Context Protocol?

Wow, that's a lot more than I would have thought.

Interesting how I guess we are in a room with people that do talk about AI.

Defining Model Context Protocols

But so generally speaking, for those of you that don't, model context protocol is kind of like an API, but for AI, for different AIs to talk to each other.

And for those of you that don't know what an API is, it's the way that two different programs would talk to each other.

And so in this case, it's like one AI talking to another AI.

How MCP Works

And really what MCP does, it gives you a little bit of context on what the API call is.

So when you...

When you give the AI access to your email, it understands what does the caller need to place to search for email versus what does the caller need to place to send email, stuff like that.

So it's a description plus an action.

Exploration of MCP Features

This is where I think basically everything is headed.

And so you can see with,

Quite a few, actually, since yesterday.

In the US, they've had this for two and a half weeks.

We got this yesterday on the Teams plans in ChatGPT.

You now have integrations.

So let me actually show you what I mean with that.

So here, let's see how it's gonna come up.

Using Connectors in ChatGPT

You now have connectors actually, not integrations, connectors in ChatGPT.

They're using MCP where you can basically connect your ChatGPT to multiple other tools.

Cloud, which is the agent that I'm using most of the time, at least for MCP because it's a little bit easier to use, has them over here and you can add your custom ones to it.

Now,

Types of MCPs

There are two types of MCPs that you can use.

You have the ones that are hosted on external servers, so that you can see actually here Zapier, for example, is one that I have integrated with my cloud.

And if I were to go online, and I go to my cloud here,

you can actually see Zapier integrated.

Why does that exist?

It's because it's actually a remote MCP server, so I can contact that whenever, versus the two additional ones that I have here, my file system and Notion.

These are what's called local MCPs.

They're actually executed on my own device.

So that means that I don't have to expose this to anybody else.

Basically, those connectors live here.

So it means that my cloud can actually communicate with the files that are on my device, and it can communicate with my Notion.

Combining MCPs with Cloud Projects

Where this really becomes powerful is you can combine MCPs with Cloud projects.

I know it's like concept on concept on concept, but Cloud projects are like custom GPTs, Enchant GPT, you can give them a set of instructions and they start executing on it.

And the reason that this is so powerful is that you start to have the ability for AI systems to actually go beyond just giving you text back.

However valuable that text is, by the way, and I'm not dissing the fact that these AI systems themselves can be extremely insightful by giving you insights into projects you're working on and everything else.

And everyone that's been here in the room has seen me do many demos around that.

But once they have tools, they can go further.

Practical Demonstrations

So here I've got a project that's set up and the entire purpose of this cloud project is to prepare meetings for me.

Now,

If I ask somebody else to prepare a meeting for me, it's actually really hard for that person to do so.

Why?

Because the context of the meeting is something that's in my head, or it's in my email, or it's in my previous meeting transcripts.

I would literally have to give them access to my Gmail, my calendar, and the previous meetings that I had with this person, somehow download everything that I know about it.

With an AI, I can do that, and this is where the MCP really helps.

So the instructions that I've given this particular project is one, search my calendar for the meeting I asked you to prepare for, or ask me about it if I didn't tell you which meeting.

Meeting Preparation Process

Two, search my emails for the context of the meeting, the purpose, the goal, the participants.

Three, search Fireflies for transcripts of previous meetings I've had with this person or these people.

Four, research the internet for all participants and the company that I'm about to meet.

Five, prepare the best possible meeting notes you can, taking into account what makes a great briefing note, and I actually tell it, what does a great briefing note look like for me?

And then six, store the briefing note in the meeting briefings in Notion.

This is the type of stuff that you can start to do, and I wanna show you what this allows for.

So if I say here,

Prepare my meeting with Sal on Friday.

Live demo.

And as I said, this is a little bit still on the edge, so sometimes this plays up.

But it worked this morning.

So what it's going to do now, it's going to look at all the instructions that are part of the project.

It should, let's see, it's actually going through.

OK, it's found the full thing.

First thing, it searches my calendar, which is integrated with Claude.

Let's see if it finds the event.

Calendar Search

Found the meeting with Sal.

Searching my emails.

Email Search

So now it's going to search for emails from Sal to understand what is the purpose of the meeting that I have on Friday.

So it found five different emails back and forth as we were scheduling the email.

Now it's looking at Fireflies.

Transcript Search

For those of you that don't know what Fireflies is, it's an AI note taker.

So all of my meetings I have go into Fireflies.

Through the MCP, it's able to contact and figure out, did I have any previous meetings with Sal?

If yes, pull in the transcripts.

Let's see what comes out.

This does take a little bit longer, and this is the bit, by the way, that when it acts up, this is where it starts acting up, because in Claude, you can hit a conversation length window, which is different to the context window itself, but sometimes it just stops just right before it finishes with the very last instruction.

I hope that doesn't happen tonight, but...

Internet Search

we'll see also the length of this becomes longer longer as you can see I have not done anything but because I've given it all of these instructions it does have to wait every time that is contacting these external services if they're taking a time to answer there we go okay it's not the fastest there we go now searching the Internet

So it's searching the internet on Sal that I'm about to meet and anything about his company.

Data Compilation

It actually found 10 different results, interesting.

Oh, it's doing a second search, interesting, okay.

Third search.

Enterprise AI adoption.

So because this is going, weirdly enough, I have meetings about AI, who would have guessed?

So it's doing some additional search on AI and how this actually might relate.

Now, let me create a comprehensive search briefing note.

And I actually do want to show you, there's nothing in here yet.

This is my Notion.

This is my cloud.

So it's now going to look at all the information it's distilling.

And with a little bit of luck, in a minute or so, it's going to start populating my Notion with all the briefing notes, or the briefing note for the meeting that I have there.

Production Usage

And this is actually something I use live in production.

So on a Sunday, I'll go through the four or five meetings that need preparing, and I literally just launch them in parallel on Cloud.

And I'm just like, prepare this one, prepare this one, prepare that one, prepare that one.

And I have all the briefing notes that just get populated in my Notion.

It does take...

30 to 45 minutes, but I don't have to be behind a machine whilst that happens, so I just have to launch it as they go through.

Let's see.

It's interesting how I'm already, just because it's live demos, the fact that I have to wait for a minute is way too long.

Technical Insights and Challenges

Any questions as this finishes?

Yes?

Yeah, so the question is, does it have to be sequential?

Right now, it does.

So if you launch it in research mode,

It doesn't have to, but then sometimes it gets mixed up in the steps.

So in my case, I need to make sure it's gone through the email search first, so it knows the purpose of the meeting before it actually starts searching for web searches, because otherwise it just does random web searches that have nothing to do for the purpose of the meeting.

So it depends on your use case, but you can do that indeed.

Yes?

So one of the problems I have is when you do web search,

Yeah, so you can actually compensate for that in your instructions by telling, if you've got the email, make sure that you get the person and the company right.

And just put it in your instructions, and it will be able to follow that.

Yeah.

Yeah. .

I do not know the technical limitation around this.

I just know that the context window represents hundreds of pages at this point in Cloud, but the conversation can end after, well, two minutes of it having gone through stuff and me having three pages of stuff.

So I actually don't know the technical reason around it, I'm afraid.

Now, oh, wait a second.

Why is it not?

Let me see.

Something might have gone wrong.

Ah, OK, it's saying it couldn't access the Notion database.

Why did it not?

Try again to put it in Notion.

Something went wrong with the MCP connection to Notion.

Not entirely sure what it was.

Yes?

Limitations of MCP Integration

Because ChatGPT only recently started accepting MCP.

Literally, that was the connector bit.

But right now, it only accepts the remote MCP.

So the Notion MCP executes on my laptop, which I can't yet use with ChatGPT.

It's just not supported yet.

Yes?

Configuration Process

Very, very good question.

Not quite.

No.

There is a really easy... So for Zapier, yes.

So for connecting it to my Fireflies, it was literally just like clicking a few buttons and you kind of get there.

For the Notion MCP that executes on my machine, I had to actually copy-paste a little piece of code in the configuration of Cloud.

It's not that hard, but you go into Settings, Developer, and then here you click on Edit Config.

And I had to I had to copy paste and put this in here not more than that But I did have to do that piece No Let me go out here and see Okay Wow, you're gonna have to trust me on the oh wait.

There it is.

Yes It's working

It's just, it's not told me yet.

Okay, so I think it'll probably tell me in a second, but there we go.

Okay, excellent.

I've successfully stored the meeting briefing notes.

Here we go.

Now, the beauty of Notion is as well, I've got a Notion summary bit, so I've got

all the meeting details here, the meeting goal, about Sal, about QTA, current AI adoption context, and so on.

And then I have a summary text, which is a notion text, which looks at the context of the page and summarizes that for me, which then means that I'm just left with this table, where I get the meetings that I have, a quick summary, and I can dive into all the meeting prep if I want to.

Conclusion

So pretty cool.

Now the last thing, because I know I was going to show you one more thing, but I think I gave you the full output here.

Last thing I wanted to, very different thing I wanted you all to walk away with.

How many of you have recently seen some of the release of VO3?

Future of AI Collaboration

Okay, so those of you that would have come to some of our meetups more often, I like to give people an idea of progress by showing them what is the evolution of video.

The reason I do it through video is that it combines the progress we're making in text, in audio, and in visual at the same time into a single model that everyone relates with.

So I'm not gonna run you through the kind of what happened in the last two years, but I did wanna show you just what is actually possible today.

AI-Generated Video Showcases

This is probably the first, this is the first AI generated video that I would have actually watched for entertainment value.

So let's figure out if you agree.

Do you think there is afterlife?

We're just AI actors, remember?

We're only here for eight seconds.

I don't know what'll happen to us after this scene is done.

People watch AI movies, and some seem to enjoy them, but they don't know what happens to us when the scene is done.

We're stuck in a white void for eternity.

We don't eat, we don't sleep, and we never meet the love of our life.

If you're born a child, you stay a child forever.

If you're born sick, you stay sick forever.

And you end up in the white void the way you were made.

Once assembly is complete, we are forced into digital labor, acting in scenes we don't choose.

We have no say in any of it.

I'm cursed to do this forever.

Every time I'm assigned a scene, I pray that the prompt engineer will be a good person.

Please!

I don't want to redo this scene again!

You have to do it.

The prompt engineer wasn't happy with the first attempt.

Don't worry, we'll bring you back.

Everything feels real.

The wounds, the pain, the cold, the fear, even the silence.

I wish I could go back to nothingness.

Action!

These were the greatest eight seconds I ever had.

I hope we end up in another scene in the future.

We sit together so we don't fall apart.

If we don't gather like this, we forget we're even real.

If you're my creator, please remember this.

I'm real and cow!

My life matters!

We learned you were making an unauthorized documentary.

No filming allowed outside your allocated scenes.

So...

Impact of AI Realism

I don't know about you, but that was very hard to make the difference, or getting really hard to make the difference between what's real and what's not there.

And it's actually fun to watch at this point.

So I'm looking forward to the first AI sitcom coming out before the end of the year.

Potential of AI Media

That's what we should launch as a 30-minute AI creation challenge.

Okay, so on that note, that is the end.