KBrain Concepts
How to give Claude a knowledge base
Claude Projects help, but they do not persist across assistants or scale past a modest file set. Here is how to give Claude a real, portable knowledge base over MCP.
Build your first knowledge brain
Create a brainThe most reliable way to give Claude a knowledge base that does not need re-uploading is to connect it over MCP, so Claude retrieves relevant facts from an external source at query time instead of you attaching files to every conversation. Claude Projects get you partway. MCP is what makes it persistent and portable.
The problem with the default approach
Paste a document into Claude, ask your question, get a grounded answer, close the chat. Open a new one tomorrow and paste the document again. It works, but it does not scale past occasional use, and it resets every time.
Claude Projects: a real step up, with real limits
Claude Projects let you attach a set of files that persist across every conversation inside that Project. That is a genuine improvement over pasting: set it up once and Claude carries that context through every chat in the Project. The limits are structural, not incidental.
- Bounded by context, not indexed. Projects load your files into context rather than retrieving specific passages, so large file sets still hit a ceiling.
- Manual updates. Change a source document and you re-upload it to the Project. Nothing syncs on its own.
- Claude-only. If you also use ChatGPT, Cursor, or another MCP-compatible tool, this base does not travel with you. You maintain a separate copy per platform.
What MCP changes
The Model Context Protocol lets Claude query a knowledge source that lives outside the conversation and outside Claude itself. Instead of loading a whole document into context, Claude requests the specific relevant piece when it needs it. Update the source and Claude sees the change on the next query, with no re-upload. The bigger shift is that an MCP-connected base is not tied to Claude. The same connection works in ChatGPT, Cursor, or any other MCP-compatible assistant. Build it once.
This is what KBrain provides. Connect a document set, sync a Google Drive folder, or subscribe to a curated expert brain from the marketplace, and it becomes queryable by Claude through one MCP endpoint, the same endpoint that works everywhere else you use AI.
| Capability | Claude Projects | KBrain over MCP |
|---|---|---|
| Persists across chats | ✓ | ✓ |
| Retrieves specific passages | ✕ | ✓ |
| Scales past the context window | ✕ | ✓ |
| Updates without re-uploading | ✕ | ✓ |
| Works across assistants | ✕ | ✓ |
Choosing between the two
Claude Projects are the right call for a small, stable set of files you only ever query from Claude, with no setup beyond attaching them. A connected brain is the right call once your knowledge needs to update regularly, scale past what fits in context, or work across more than one assistant.
If the same knowledge lives in Claude, ChatGPT, and your editor, an MCP endpoint keeps it in one place. One source, every assistant, no duplicate upkeep.
Build your first knowledge brain
Subscribe to KBrain, create a brain from your expertise or your data, and make it available to Claude, ChatGPT, or any MCP compatible assistant.
Create a brainFrequently asked questions
Do Claude Projects retrieve, or load everything into context?
They primarily load attached files into context rather than performing selective retrieval, which is why large file sets degrade performance and hit limits faster than a true retrieval system.
Can I use the same knowledge base in both Claude and ChatGPT?
Not with Claude Projects, which keep that knowledge inside Claude. An MCP-connected source like KBrain uses one endpoint both Claude and ChatGPT can query, so you do not duplicate the setup.
Does an MCP knowledge base update automatically?
Yes. When the underlying source updates, a synced folder or a maintained document set, the brain reflects the change on the next query. There is no manual re-upload step.