So you have a cool AI tool you love to use. What if you could easily plug it into your own private knowledge base, built with CustomGPT.ai?
Refer to our Other clients + CustomGPT.ai’s Hosted MCP server docs for more info.
If your tool supports the Model Context Protocol (MCP), you can. MCP is an open standard, which is a fancy way of saying it’s a “universal plug” for AI. This post will give you the general idea of how to connect any MCP-friendly tool.
The Big Idea: MCP
MCP is just a set of rules that lets an AI app talk to a data source. When a tool says it “supports MCP,” it means it knows how to use this universal plug. This lets you connect your CustomGPT.ai agent—with its top-ranked accuracy for finding answers in your docs—to the tools you already use.
Want to geek out on the details? Our Hosted MCP launch blog post has you covered.
The Two Ways to Connect
Most tools will connect in one of two ways.
1. The Direct Connection: The app talks straight to the CustomGPT.ai server. It’s simple and fast.
Your AI Tool --> (Securely over the Internet) --> CustomGPT.ai Server --> (Back to your tool)2. The Helper Connection: The app uses a little helper program on your computer (called supergateway) to manage the connection.
Your AI Tool --> Your PC's Helper --> CustomGPT.ai Server --> Your PC's Helper --> Your AI ToolYour tool’s documentation will tell you which method it uses.
What you’ll always need from CustomGPT.ai:
No matter how you connect, you’ll need to grab a few details from your CustomGPT.ai dashboard.
- A CustomGPT.ai account and a project with your data.
- Your Project ID.
- Your MCP Token (which is like a secret password).
You can find all of this by logging in, opening your project, and clicking on the Deploy ➞ MCP Server (Beta) tab.
The General Steps
- Find the MCP settings in your tool. Look for “MCP,” “External Context,” or “Add Server” in your tool’s settings or preferences.
- Provide your details. Your tool will ask for your server info.
- If it’s a Direct Connection (like Cursor or Windsurf), it will probably ask for a single URL. You’ll use this format:
https://mcp.customgpt.ai/projects/<PROJECT_ID>/sse?token=<TOKEN>- If it’s a Helper Connection (like Claude or Trae), it will ask for a command and arguments, which you can copy directly from the JSON block in your CustomGPT.ai MCP dashboard.
- Save and go!
That’s the basic pattern. By using this standard, you can turn CustomGPT.ai into the central brain for all your favorite AI applications.
Watch what’s possible:
Your Turn to Build!
So there you have it. No more generic answers! Your favorite AI tool is now your personal expert, with a direct line to the documents and data you care about most. It’s a true game-changer for getting things done.
Why not give it a try? Head over to CustomGPT.ai to start a free trial and create your own private AI brain.
We can’t wait to see what you build with it!
P.S If are into MCP world, you will definitely live this MCP AMA that we did with Pinecone folks and other MCP folks (Santiago).
Find it here:
Further Reading
For those interested in exploring MCP in greater depth, check out the following resources:
- What Is Model Context Protocol (MCP)?
A comprehensive introduction to MCP’s origins, goals, and why it’s becoming the de facto “USB-C” for AI integrations. - Top 124 MCP Servers, MCP Clients & Tools (2025 Guide)
A curated catalog of production-ready MCP servers and clients—both open-source and proprietary—showcasing plug-and-play tools you can use today. - Problems MCP Model Context Protocol Solves
An exploration of the real-world pain points (API sprawl, ad-hoc integrations, security gaps) that MCP was designed to eliminate. - The Model Context Protocol (MCP) Architecture (2025)
A deep dive into MCP’s component layers, sequence diagrams, security considerations, and production-grade deployment patterns.
Frequently Asked Questions
Will any AI model be trained on my data if I connect through MCP?
No. The provided compliance credentials state that data is not used for model training. For MCP access, you connect an MCP-compatible tool to a specific project using a Project ID and MCP token, and the service is listed as GDPR compliant and SOC 2 Type 2 certified.
Why connect an AI tool to an MCP knowledge base instead of wiring the tool straight into our business systems?
Stephanie Warlick, Business Consultant, describes the value this way: u0022Check out CustomGPT.ai where you can dump all your knowledge to automate proposals, customer inquiries and the knowledge base that exists in your head so your team can execute without you.u0022 You would typically use an MCP knowledge layer when you want multiple AI tools to read from one reusable source of truth. For programmatic workflows or broader system automation, the platform also supports an OpenAI-compatible API and 1400+ Zapier integrations.
Do I need to host my own MCP server, or can I use a remote hosted one?
You do not need to host your own server for the documented setup. The supported flow uses a hosted MCP server in two patterns: a direct connection from the AI tool to the remote server, or a helper connection where a local program such as supergateway forwards requests. Bill French, Technology Strategist, said, u0022They’ve officially cracked the sub-second barrier, a breakthrough that fundamentally changes the user experience from merely ‘interactive’ to ‘instantaneous’.u0022
Which AI tools can actually connect to a hosted MCP server?
An AI tool can connect if it supports MCP or gives you a place to add an external server. The documented examples are Cursor and Windsurf for direct connections, and Claude and Trae for helper-based connections. A practical test is to check the tool’s settings for labels such as u0022MCP,u0022 u0022External Context,u0022 or u0022Add Server.u0022
How easy is it to load my documents before using MCP, and will a PDF of a spreadsheet work?
Evan Weber, Digital Marketing Expert, said, u0022I just discovered CustomGPT, and I am absolutely blown away by its capabilities and affordability! This powerful platform allows you to create custom GPT-4 chatbots using your own content, transforming customer service, engagement, and operational efficiency.u0022 In practice, you load your source material into a project first, then your MCP-compatible tool queries that indexed knowledge. Supported formats include PDF, DOCX, TXT, CSV, HTML, XML, JSON, audio, video, and URLs, with a 100MB limit per file. If your spreadsheet is already available as CSV, that format is also supported.
When should I use the hosted MCP server instead of the RAG API?
Use the hosted MCP server when you already have an MCP-aware AI tool and want to connect it to your project with minimal custom development. Use the OpenAI-compatible REST API at /v1/chat/completions when you need to build your own interface or integrate programmatically with SDKs for Python, Node.js, .NET, Java, Go, PHP, Ruby, or Swift. Barry Barresi, Social Impact Consultant, described one tailored implementation this way: u0022Powered by my custom-built Theory of Change AIM GPT agent on the CustomGPT.ai platform. Rapidly Develop a Credible Theory of Change with AI-Augmented Collaboration.u0022
Related Resources
This guide pairs well with a deeper look at how CustomGPT.ai extends agent capabilities.
- Custom Actions Guide — Learn how to connect external tools and services to CustomGPT.ai so your agents can take action beyond answering questions.
Priyansh is Developer Relations Advocate who loves technology, writer about them, creates deeply researched content about them.