YouTube Deep SummaryYouTube Deep Summary

Star Extract content that makes a tangible impact on your life

Video thumbnail

How to Build a Unified Ecosystem inside Cursor AI

AI LABS β€’ 2025-06-08 β€’ 10:03 minutes β€’ YouTube

πŸ€– AI-Generated Summary:

Streamline Your AI Workflow with the ACI Unified MCP Server: A Game-Changer for Cursor and Beyond

If you're a developer or AI enthusiast using Cursor or similar AI coding assistants, you know how powerful it is to extend functionality by adding Multiple Custom Plugin (MCP) servers. Whether it's the Figma MCP, GitHub MCP, or Firecrawl MCP for web scraping, these plugins can significantly boost your productivity. However, adding too many MCPs directly inside Cursor can lead to clutter, performance slowdowns, and context loss.

In this post, we dive into an innovative solution that solves these challenges β€” the ACI Unified MCP Server from aci.dev. This powerful platform centralizes and manages your MCP integrations, offloading the heavy lifting from Cursor’s own agent and dramatically improving responsiveness and stability.


The Problem with Multiple MCPs in Cursor

Adding multiple MCP servers inside Cursor can quickly become overwhelming for your system, especially if your hardware isn’t powerful enough or the MCPs are running locally rather than on external servers. Each MCP executes commands on your machine, and with many plugins active simultaneously, Cursor’s performance can degrade significantly.

From personal experience running six MCPs inside Cursor, the app became slow and choppy during AI code generation. Removing extra MCPs restored performance immediately, highlighting the need for a better MCP management approach.


Introducing the ACI Unified MCP Server

The ACI Unified MCP Server acts as a single AI agent that manages all your MCP servers in the background. It automatically selects the most relevant MCP server based on your task and executes commands externally, keeping Cursor lightweight and responsive.

Key Functions:

  • Search function: Efficiently search across integrated apps.
  • Execute function: Automatically pick and execute the correct MCP for your request.

Because the ACI MCP server handles all the decision-making and processing, your Cursor or Windsurf agent won’t slow down or lose context, regardless of how many MCPs you use.


Getting Started with ACI Unified MCP Server

1. Explore the Platform

The ACI platform is fully open source and currently supports over 600 tool integrations that you can run locally without sharing your data. The platform is rapidly evolving with new apps added frequently, including upcoming integrations like Google Drive and Google Calendar.

2. Configure Your Apps

To use apps with the MCP server, you need to configure them on the ACI dashboard:

  • Log in to the aci.dev website.
  • Access the App Store and select apps you want to use.
  • Configure each app by providing necessary authentication credentials (e.g., API keys).
  • Some apps require no authentication and can be used immediately.
  • Assign a linked account owner ID (e.g., "AI Labs") to group your app configurations.

3. Set Up Agents and Link Apps

  • Create or select an agent in the dashboard.
  • Link your configured apps to the agent.
  • This agent will be used by the MCP server to handle requests.

4. Add the Unified MCP Server to Cursor or Windsurf

  • Download the MCP JSON configuration from ACI documentation.
  • Replace the placeholder variables:
  • Linked account owner ID (the one you created, e.g., "AI Labs").
  • ACI API key (found in your agent dashboard).
  • In Cursor or Windsurf, add a new MCP server using this JSON config.

Enhancing Your Cursor Experience

Add a Custom Prompt Rule

To improve interaction quality and reduce hallucinations, add the provided ACI prompt text into Cursor’s Project Rules. This prompt informs the agent about the MCP server’s capabilities and optimal usage.

Testing the MCP Server

Once set up:
- Ask your agent to list all available MCP apps.
- Try executing functions like searching research papers or querying Brave Search API.
- The agent will route your request through the unified MCP server, returning accurate and relevant results.


Troubleshooting Authentication Issues

If you encounter authentication errors (e.g., β€œuser isn’t authenticated”):

  • Open the problematic app’s configuration on the ACI dashboard.
  • Add a new linked account owner ID.
  • Update your MCP JSON configuration to use this new ID.
  • This simple step resolves rare credential mismatches.

Why Use the ACI Unified MCP Server?

  • Improved performance: Offloads MCP management from Cursor’s agent.
  • Simplified configuration: Manage all your MCPs in one place.
  • Scalability: Add as many MCP integrations as you want without slowing down your system.
  • Privacy: Runs locally with no data sharing required.
  • Open source: Transparent and customizable platform.

Final Thoughts

The ACI Unified MCP Server is a real game-changer for anyone using Cursor, Windsurf, or similar AI coding tools. It solves the common bottleneck of managing multiple MCPs, freeing your AI assistant to perform with speed and accuracy.

If you want to enhance your AI coding workflow and try out this powerful tool, head over to aci.dev, explore the integrations, and start configuring your unified MCP server today.


Support the Creator

If you found this guide useful, consider subscribing or supporting the channel through memberships or the super thanks button. Your support helps create more tutorials and keeps the community growing.


Stay tuned for a deeper dive into the ACI platform and its exciting capabilities. Until then, happy coding and MCP managing!


Helpful Links:
- ACI Unified MCP Server Documentation
- Cursor AI
- ACI App Store


Feel free to share your experiences or questions in the comments below!


πŸ“ Transcript (291 entries):

I'm here in cursor settings going into the MCP tools. This is where you see all of your MCPs added. Nowadays, if you want to be really proficient and enhance your workflow, you can add a lot of MCPs. Whether it's the Figma MCP, the GitHub MCP, or the Firecrawl MCP for web scraping. You can add all of them right here in your MCP tools. But when you add too many of these MCP servers, the agent becomes cluttered. It gets harder for it to keep track of all the MCPs and ultimately decide which one to choose. If you don't have a powerful enough system and those MCPs aren't running on an external server, they actually run on your own machine. Every time you refresh them, they execute a command that runs on your system. Believe me, I had six of these MCPs running right here in cursor. And cursor itself became really slow and choppy. Whenever the AI was generating code, performance dropped significantly. The only reason I could think of was the overload from adding all these MCPS. Sure enough, when I removed them, cursor went back to normal. Today, what I really want to talk about is this amazing unified MCP server from aci.dev. It does something pretty cool. It has two main functions. The search function and the execute function. You can think of this MCP server as a separate AI agent that has all the other MCP servers running in the background. It automatically chooses the right one based on your task and executes it. That entire load is lifted off Cursor's own agent. Now, Cursor won't slow down. It won't lose context. and you won't have to manually add multiple MCP servers inside Cursor. This is a real gamecher and I'm going to show you exactly how you can get all your MCP servers linked through the ACI Unified MCP server. If you like the content we're making, I'd really appreciate it if you could subscribe. Right now, we're also testing out memberships to support the channel. We've only launched the first tier and for now it offers priority comment replies, but subscribing helps us see how many of you are interested and want to support what we're doing. This is the platform and the platform itself is a whole other topic. It's an excellent system and I'll definitely be doing a separate video to dive deeper into it. For now, I want to guide you through how to set up your unified MCP. Just a heads up, this platform is fully open source. You can basically use it to create tool integrations and there are currently over 600 of them that you can run on your own local computer. You don't need to share any data at all. That's pretty powerful and I'll definitely be doing a deep dive into it later. Next, log in to the website and you'll be taken straight to their dashboard. Once you're in, you'll be greeted by the app store. There's a lot of variety here. If you think some apps are missing, don't worry, they'll be added soon. For example, if you want to build something with Google Drive integration, support for that is also on the way. This platform is growing really quickly. When I first discovered it, it only had around 9 or 10 integrations. Since then, a lot more have been added and even more are on the horizon. First, you need to configure an app for yourself so you can use it in the agent playground or even with your MCP servers. Wherever you want to use the app, you'll need to come into the app store and configure it. Let's say I want to use the Brave Search app. I'll just open it and set it up. As you can see, I've already done that. For all apps, configuration is essentially a way to authenticate with an external service, but they're not all the same. For Brave Search, I just had to get my API key, and that was all it took. Now, take something like the archive platform, which is a place where research papers are uploaded, a full collection of them. Here you can see we have various functions. Each app has its own set of functions, which are essentially tools for that app. for example, search papers, get paper metadata, get daily updates, and more. Let's say we want to configure this app. It's pretty straightforward. There's no authentication method required since it's open- source. So, we can just use the APIs directly. Next, you have to select an agent. If you've set up multiple agents in your API, you can choose which one to link this app to. Right now, I just have one. Then, there's the account. As you can see, there's nothing here. If I were integrating the Brave Search API, there would be another box where I'd add the API key. But since nothing is required here, I'll just add my account owner ID, AI Labs, and save it. Now, the account has been linked successfully. To show that these apps are now available, let's go into the agent playground, which is where we actually build agents. I have only one agent, so I'll select the default agent. Then choose the account owner ID, AI Labs, then apps. This is the second one. Here you can see we have two apps set up. Let's say I choose this app to be accessible by the agent. Then in the function section, I can pick whichever function I want. They're all available. This is how you configure apps on this platform. For something like Google Calendar, the integration is a bit more complicated. You'd have to go into your account, get all the credentials, and fill them out. But there's another option. You can use ACI Dev's own authentication, which is just for testing. If you actually build an agent from this platform, you'll still need to use your own credentials. So, it's best to stick with the default configuration and fill out the required fields if you want to configure any more apps. All the apps you've configured will be available right here in the app store. Just come here and check out the ones you've already set up. If you want to configure more, you can do that right here as well. They provide a pretty good guide. You just need to figure out where to get the credentials for those more complicated app connections like I showed you earlier. Other than that, it's really straightforward. Next, go into linked accounts, and you'll see that all these apps are connected to a linked account owner ID. I've kept it the same for all the apps I've added. That's important because the same ID will be used in the MCP server connection we'll be setting up. The rest of the process for adding the MCP server to Cursor or Windsurf or Claw Desktop, whichever app you're using, is actually pretty easy. It's not complicated at all. The only part that might feel tricky is if you're setting up an app configuration that requires a lot of credentials. Other than that, it's a very straightforward process. Right now, I'm in the ACI documentation. I'll drop a link for this in the description below. Scroll down to the MCP server section, and you'll find the unified MCP server. In the introduction, they explain that there are two types of MCP servers. One is the apps MCP server, and the other is the unified MCP server. To give you a quick idea, the app's MCP server exposes all the apps at once just like cursor would. It keeps all the MCPs offloaded from cursor, but the agent still has to decide which one to use. The unified MCP server, on the other hand, provides functions that actually help the agent, whether it's Cursor's agent or Windsurf's agent, automatically choose the most relevant MCP server based on the task. Scroll down to the unified MCP server section and you'll find a section called integration with MCP clients. This includes different clients like cursor, windsurf, claw desktop or even local environments. Open the cursor and windsurf section and you'll see an MCP JSON file. This is the format used to add MCP servers to Windsurf and cursor. The entire configuration is already provided. You just need to replace two variables. One is the linked account owner ID which in our case is AI labs. You would have created this ID when setting up your app connections. So just use that same ID. The second is your ACI API key. To get this, go back into your ACID dev dashboard and head to the agent section. Here you'll see we've been talking about the default agent. This is the one the MCP server will actually use. The default agent has an API key that allows it to be accessed. Copy this key and paste it into your MCP JSON configuration. Now head back into cursor and go to add a new MCP server. open it and you'll see that I've added my ACI MCP unified server along with the API key and the account owner ID AI labs. Now the ACI MCP unified tool is available to us. Expand it and you'll see that all the tools inside it are available as well. Another thing they've asked us to do is add this prompt which tells the cursor agent what the MCP server is about, what functions it has, and how to use them properly so we get a better experience. This prompt isn't mandatory. It's simply meant to improve the experience and help the agent use the MCP without any disturbances or LLM hallucinations. If you're worried about having to paste the prompt repeatedly, don't worry. Just copy it, head back into cursor, and go to the rules section. There you'll find project rules. Simply add a project rule and it will automatically be included in the context every time. So, you won't need to manually add it again and again. Just to show you how the MCP is actually working, I went ahead and tried it out. First, I pasted the prompt they asked us to use, and the agent responded saying it understood how to use the MCP. Then, I asked if it could list all the MCP apps I had access to. There really wasn't any room for confusion since I had disabled all the other MCPS and provided a clear prompt. The agent called the ACI search functions MCP tool. And as you can see, we gave the command in natural language. That request went through and the agent returned our apps like the archive app and the Brave Search app. It even listed the available functions like get category search papers and we can use those functions directly right here. I also searched the brave search API for the tallest building in the world and as you can see the tool call went through. The query was what is currently the tallest building in the world using the Brave web search tool. It returned the correct result. Burge Khalifa is the tallest building. If for some reason you start receiving weird authentication errors like it says your user isn't authenticated, here's what you can do. Go into the app configuration for the app that's giving you the error. For example, if it's happening with archive, open that app, add a new account, and create a new owner ID. Save it, delete the previous one, and make sure to replace the owner ID in your MCP configuration. This is just a rare issue we encountered once during testing. It didn't happen again, but if it does happen to you, simply change the linked account owner ID name and you'll be good to go. That brings us to the end of this video. If you'd like to support the channel and help us keep making tutorials 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.