KBrain Concepts
How to connect a knowledge base to ChatGPT via MCP
A step-by-step look at connecting an external knowledge source to ChatGPT with MCP, and how it compares to custom GPTs and file uploads.
Connect your knowledge over MCP
Connect via MCPTo connect a knowledge base to ChatGPT with MCP, you add the source as an MCP connector in ChatGPT's settings using the endpoint your provider gives you, then enable it for the conversation. Once connected, ChatGPT queries that source directly instead of relying only on what was pasted or uploaded.
Where this fits versus custom GPTs
ChatGPT already supports retrieval through custom GPTs with file search, which is a legitimate option if you only ever use ChatGPT and your knowledge set does not change often. MCP connectors solve a different problem: portability. A custom GPT's knowledge lives inside that GPT, inside ChatGPT. An MCP-connected source is external, so the same connection works in ChatGPT, Claude, or any other MCP-compatible client. You build it once instead of once per platform.
inside ChatGPT
every MCP client
The general setup
The menus shift as OpenAI updates the interface, but the process keeps the same shape.
- Get the MCP endpoint from your knowledge provider. KBrain generates one automatically for any brain you create or subscribe to.
- Add it as a connector in ChatGPT's settings, under connectors or MCP servers, depending on the current interface.
- Authorize access if the source is private. Marketplace brains may not require this step.
- Enable the connector for your conversation. Some ChatGPT interfaces require this per chat, not just once globally.
- Query normally. Ask something the connected source covers, and ChatGPT retrieves the relevant passage while forming its answer.
What changes once it is connected
- Persistent, not per session. The knowledge is available in every future conversation, not only the one where you set it up.
- Retrieval instead of context-stuffing. ChatGPT pulls the specific relevant piece rather than you loading a full document into the chat window.
- Cross-platform by default. The same MCP endpoint that works in ChatGPT also works in Claude and other compatible assistants, with no separate setup per tool.
Common issues
- ChatGPT is not pulling from the connected source. Confirm the connector is enabled for the active conversation. Being connected in settings does not always mean it is active in every chat.
- Access is denied. Private brains require authorization from whoever created them. To query someone else's private source you have to be granted access first.
- Answers still read generic. MCP retrieval only helps when the query falls within what the connected source covers. Outside that scope, ChatGPT falls back to training-based reasoning, which is the behavior you are trying to avoid on anything specific to your domain.
Support for MCP connectors has rolled out by plan tier and changed over time. Check OpenAI's current documentation for what your plan supports before assuming access.
Connect your knowledge over MCP
Create or subscribe to a brain on KBrain, then connect its MCP endpoint to Claude, ChatGPT, or any MCP compatible assistant in a few minutes.
Connect via MCPFrequently asked questions
Is MCP better than a custom GPT for knowledge retrieval?
Not strictly better, different tradeoffs. A custom GPT is simpler if you stay entirely inside ChatGPT. MCP is the better call when you want the same knowledge base in more than one assistant, or when the source needs to stay current without manual re-uploads.
Does every ChatGPT plan support MCP connectors?
Support has rolled out by plan tier and changed over time. Check OpenAI's current documentation for what your specific plan supports before assuming access.
Can I connect multiple knowledge sources to ChatGPT at once?
Yes, and ChatGPT uses whichever is relevant to the query. Scope each connector to a clear domain rather than overlapping sources, which keeps retrieval precise.