The MCP knowledge base for Claude and ChatGPT
An MCP knowledge base
your AI can read and write.
Hjarni is a clean knowledge base with a built-in MCP server. Claude and ChatGPT read your notes, follow your instructions, and write new ones back. All in plain Markdown they read perfectly, instead of choking on a giant paste.
Free to start. No credit card required.
The problem
Stop pasting. Start connecting.
Every new chat starts from zero, so you paste. Your stack, your decisions, the same Markdown and config files, over and over. And half the time the model chokes on the dump and answers from generic training data anyway.
An MCP server fixes that. Your AI connects to your knowledge base and reads your notes cleanly, in the format it already speaks. No giant paste. It reads what it needs, when it needs it.
Hjarni is that knowledge base. The MCP server is built in, not bolted on.
How it works
Save notes once. Connect with MCP.
Two steps. Then every conversation starts from your context instead of a blank page.
01
Write it down
Notes in plain Markdown, organized in folders and tags. The format your AI already speaks.
02
Connect the MCP server
Add Hjarni once in Claude or ChatGPT. Sign in. That is the whole setup.
03
Your AI reads and writes
Claude and ChatGPT search your notes, follow your instructions, and save new notes back to the right folder.
https://hjarni.com/mcpStep-by-step for Claude, ChatGPT, and every other client. Check a config with the MCP config validator.
Why this MCP
A toolset your AI won't get lost in.
Light and well defined. Search, read, create, update. Just enough for your AI to use it well, with no sprawling surface area to wander into.
Read and write
Search, read, create, and update notes. Your AI is not a read-only guest. It keeps your knowledge base current.
Markdown end to end
Notes go in and come out as Markdown. No conversion, no lossy formats, no lock-in. Export everything anytime.
It carries your rules
The server hands your per-folder instructions to whichever AI is reading, so it writes notes that fit your system.
I wanted a cloud knowledge base I could use as memory for my AI, in plain Markdown, with read and write access. Nothing else fit. Hjarni did. What sold me was how light and well defined the MCP is. My AI used to choke when I pasted Markdown and config files into a chat. It reads them perfectly from Hjarni. The folder-level AI instructions and the knowledge graph are the parts I did not know I needed.
Works with both
Claude and ChatGPT. One knowledge base.
Connect the same MCP server to every AI tool you use. No copy-paste between them. No vendor lock-in.
Also Cursor, Claude Code, Windsurf, Cline, Copilot, and more. See every integration.
What it isn't
Not another chatbot bolted onto notes.
ChatGPT memory
Lives inside ChatGPT, opaque, and ChatGPT only. You cannot read it, edit it as Markdown, or let Claude use it. Compare Hjarni vs ChatGPT memory.
Claude Projects
Files uploaded to one project, Claude only, no MCP, no API. Your ChatGPT cannot see them. Compare Hjarni vs Claude Projects.
A bare MCP wrapper over a database
A protocol endpoint is not a knowledge base. You still need somewhere to write, organize, and read your notes as a human. Hjarni is both.
Hjarni
A real knowledge base you read and write as Markdown, plus a built-in MCP server and full REST API. Same notes in Claude, ChatGPT, and every MCP client.
Common questions
Questions about an MCP knowledge base
What is an MCP knowledge base?
A knowledge base with a built-in MCP server. Instead of pasting context into the chat, Claude and ChatGPT connect to it and read and write your notes directly. Hjarni is one: plain Markdown notes, folders, and tags, with the MCP server built in.
What MCP server should I use for a knowledge base?
One that reads and writes, speaks Markdown, and works with every client. Hjarni's MCP server is built in, not a wrapper over a database. It has a small, well defined set of tools your AI uses well, and the same server works with Claude, ChatGPT, Cursor, and any other MCP client.
Can ChatGPT and Claude write to my notes, or only read?
Both. Your AI can search and read notes, create new ones, update existing ones, add tags, and file notes into folders. You can constrain that with folder-level AI instructions so it writes notes that fit your system.
How is this different from ChatGPT memory or Claude Projects?
ChatGPT memory lives inside ChatGPT and only ChatGPT can use it. Claude Projects keeps files inside one Claude project. An MCP knowledge base sits outside any one product, as Markdown you own, and every AI you connect reads the same notes.
Do I need a paid plan to connect an MCP server?
Hjarni's free tier includes full MCP access with no token limits. On the AI side, MCP connectors may require a paid Claude or ChatGPT plan. Check the current pricing for your client before you build a workflow around it.
Give your AI a memory. Today.
Save your first note and connect the MCP server in two minutes. Free to start.
Free to start. No credit card required.