Over the last year, AI technology has advanced incredibly fast. We've seen some amazing breakthroughs happen. Looking back at our channel, we started with a simple cursor tutorial that had a decent AI agent. But now, with new tools like MCPs and better ways to handle context, everything has completely evolved. Today, I'll show you the first platform that truly demonstrates this evolution. A platform that has taken AI agent architecture to the next level. Now, the platform I'm talking about is aci.dev. It allows you to build accurate AI agents even faster. What's powering all of this underneath is the MCP architecture they've built. Here's how it works. You create agents, select the apps you want, and then assign those apps to your agents. That's it. You now have amazing and reliable agents that actually work autonomously, and it's incredibly easy to set up. There are basically two main ways you can use this in your daily life. First, there's the platform itself. Second, they also offer two MCP servers. I've already done a video on the unified MCP server, but I'm going to give you a quick overview. Again, not many people know about it yet, and it's just insanely powerful. When you first log into the platform, you're going to be greeted with this app store. Essentially, it contains all the integrations available to you, and it's a really extensive list that you can choose from. These integrations are well documented inside the platform. So, you just need to connect them to get started. I'll show you the rest of the process in detail, but once connected, the AI agent will be able to use these tools, break down your tasks, and intelligently choose whatever tool it needs to complete your requests. After exploring the app store, we move to our app configuration section. Here you'll find all the apps that you've successfully installed and connected. You can see that right now I have three apps installed and ready to use. Then we have our linked account section, which shows all your connected services. And after that, we have our agent playground where you get this chat GPT- like interface that's really intuitive to use. On the right side of the screen, you can configure the apps you need and choose the specific functions from those apps that you want your agent to have access to. So that gives you the basic overview of how the platform is structured and organized. Now, let's move forward and I'm going to show you exactly how you can put this to work. First, we'll explore how you can use it with the agent playground feature. Now, how do you use the AI agent builder in this agent playground? First, you'll select the integrations you want your AI agent to access from the app store. This configuration step is necessary for both the AI agent playground and the unified MCP server. There are three types of apps you can integrate. The first type like archive doesn't require authentication. When I configure it, there's no authentication method needed. I just confirm it, choose the default agent, and add the owner ID. I've set mine to AIABS. After saving it appears in my app configurations because the backend API doesn't require authentication. The second type like Brave search requires an API key. Brave search lets you search the web by pasting your API key and providing a query. When configuring it, you'll see the API key requirement. After confirming, selecting the agent, and choosing the owner ID, I add my API key and save it. It now appears in both app configurations and linked accounts under AI labs. This gives you control to build different AI agents with different configurations which is powerful. The third type includes apps like Google Calendar, YouTube, Google Maps, and Gmail. These require OOTH authentication. For Google Calendar, we configure the app and fill out the required fields. If you're planning to build your own agents by cloning this open-source platform, you'll need to implement OOTH yourself. For now, we can use ACI Dev's built-in OOTH app. After confirming and adding the owner ID, we start the OOTH flow and complete the authentication. Google Calendar then appears in both app configurations and linked accounts under AI labs. Once your apps are connected, you can start using them. Select the agent. Choose the account ID and pick which apps this agent can access. I'm selecting Google Calendar and adding all available functions for this demonstration. Now I'll ask, do I have any meetings today? You can see the agent calling the function and it correctly shows I don't have any meetings scheduled for today. You can also see that if you test it further, I can ask it to set meetings with others. You'll see it's calling the event insertion function and it has successfully inserted a meeting. I actually entered the wrong email address. So, let me correct that and see if it can edit the meeting. And you can see what's happening. It's actually deleting the original meeting and inserting a new one with the correct information. Now, if I ask it, what meetings do I have? It can show me that we have our upcoming meetings scheduled. This is the kind of power this platform gives you. And it's all because it's leveraging these MCP integrations to create truly functional AI agents. Oh, and if you're enjoying the content we're making, I'd really appreciate it if you hit that subscribe button. We're also testing out channel memberships. Launched the first tier as a test, and 82 people have joined so far. The support's been incredible, so we're thinking about launching additional tiers. Right now, members get priority replies to your comments. Perfect if you need feedback or have questions. Now, you might remember that I told you this platform is faster than other agentic platforms because it uses MCPs. But here's the thing. MCPs are actually really slow by themselves. So, how does it speed them up and still achieve this impressive performance? The answer lies in their use of a vector DB. Inside this vector DB are all the descriptions of the MCP tools available on the platform. What makes this powerful is that it performs semantic search instead of plain textual search. This leads to much faster retrieval of tool descriptions and the system can more accurately determine which tool it needs to call for any given task. This entire process essentially represents a rag system retrieval augmented generation. That's exactly why it can call those tools so efficiently. It retrieves the tool descriptions more quickly and accurately than other AI agent platforms currently available. Because it can do this, the system has much more context available when building your workflows. For example, if you're asking it to check your email and schedule something on your calendar, it already has all the relevant MCP tool descriptions stored in its database. Using semantic search, it finds them faster and with greater accuracy. Once it gets that context, the AI agent can make significantly better decisions, build more accurate workflows, use exactly the right tools, and ultimately complete your tasks more effectively than traditional approaches. Now, there is an important consideration to keep in mind. If you explore the app store, as I mentioned earlier, when you configure apps with OOTH using their built-in OOTH system, you're essentially giving this platform access to your personal data, so they can still access your information. It's not a completely private solution regardless of how powerful it is. However, you already know that this platform is open- source. So, what you can actually do is clone the repository and implement it yourself on your own infrastructure. There would be some complications with this approach. For example, you'd need to set up your own credentials and configure your own ooth applications. And honestly, it's not the easiest process to navigate. It can be quite involved, especially when you're dealing with platforms like Google Cloud Platform and setting up all the necessary authentication flows. We actually mentioned this challenge in our previous video where we explored MCP in detail. If more people are interested in this self-hosted approach, we can create a comprehensive guide for you showing exactly how to integrate each tool locally into your own workflow and maintain complete privacy over your data. Now remember the unified MCP server I mentioned for developers who want to integrate this with their existing MCP clients like claude desktop cla code or cursor the setup is straightforward. There are two server types available. The apps MCP server and the unified MCP server. The unified version is more powerful and frankly that's the one worth using. The integration happens on ACI.dev using the same setup from the agent playground. Here's the key difference. The unified MCP server doesn't directly call tools. Instead, it communicates with your AI agent, which decides which tool to use. Combined with their rag system, this makes everything significantly faster. This solves two major cursor limitations. First, cursor can't handle more than 40 tools effectively. Second, multiple MCP servers create a clunky interface that slows everything down. Cursor lacks the vector-based rag system that ACI provides. Here's the setup process. Get your API key from aci.dev's agent section for the specific agent you want to use. This multi- aent support is powerful. Imagine having a coding agent and a business agent with different tool sets. Configure your apps as I showed earlier. Set your linked account owner ID. Mine is AIABS. Then integrate with your preferred client. For cursor, copy the JSON configuration, paste it in, remove unnecessary fields, and enter your credentials, the owner ID and API key. You'll see two tools appear. search and execute functions that the agent uses based on your requests. For example, asking what tools can I use right now returns Google calendar, Brave search, and YouTube since my agent has access to them. For optimal performance, use the context form from their documentation. Set it in your project rules as always to agent so cursor always knows how to interact with the MCP server effectively. And as you can see right here, it correctly identified that my next scheduled meeting is on Saturday with the email address I specified earlier. It provided me with a complete summary of the meeting details and also included the meeting link for easy access. So now you can visit the platform and start exploring these capabilities for yourself. I'm not going to fully cover the local setup in this video since it's quite extensive and would require its own dedicated tutorial. However, you can definitely try out the hosted setup right away. If you want, you can set up agents directly on the platform and use them there. Alternatively, if you'd rather integrate them into your existing workflow, you can use the MCP server I just demonstrated and start incorporating these powerful MCP servers into your daily development routine using this platform. Either way, you now have access to a much more efficient and capable AI agent system that can truly transform how you work with multiple tools and services. That brings us to the end of this video. If you'd like to support the channel and help us keep making videos like this, you can do so by using the super thanks button below. As always, thank you for watching and I'll see you in the next one.