Why most "knowledge bases" are read-only
When people connect ChatGPT to their notes, it usually reads but cannot write: file uploads, a static export, or a search-only tool. The moment your notes change, the AI is out of date again. A knowledge base ChatGPT can write to closes that loop. ChatGPT does not just answer from your notes, it keeps them current.
How writing works over MCP
Hjarni's built-in MCP server exposes a small set of tools to whichever assistant you connect: search, read, create, update, tag, and link. So ChatGPT can save the summary of a conversation as a new note, append a decision to an existing one, or tag and file it into the right folder, all during the chat. Claude, Cursor, and Claude Code use the same tools on the same notes.
Set it up in three steps
01
Create a free Hjarni account and write one note.
02
Add Hjarni as an MCP connector in ChatGPT and authorize once.
03
Ask ChatGPT to save or update a note. It writes back to your knowledge base.
Step-by-step: connect ChatGPT and connect Claude.
Keeping write access safe
You authorize the connection with OAuth and can revoke it in one click. You can also shape what gets written with folder-level AI instructions, so notes land in the right place and follow your conventions. And because every note is versioned, you can see exactly what changed.
Read-only notes go stale. A knowledge base ChatGPT writes to stays current, and Claude reads the same updates.