What is MCP?
MCP stands for Model Context Protocol. It's an open standard that lets AI assistants connect to external tools and data. Think of it as USB for AI.
Why it exists
AI assistants are powerful. But they're isolated. ChatGPT can't read your notes. Claude can't check your calendar. Every conversation starts from scratch on your data, because the AI has no way to reach it.
MCP fixes that. It's a protocol — a shared language — that lets AI assistants talk to external services. One standard. Many connections. Any AI that speaks MCP can connect to any tool that speaks MCP.
How it works
MCP has three parts:
MCP Client
The AI assistant. ChatGPT, Claude, or any app that speaks MCP. It sends requests like "search for notes about deployment" or "create a new note."
MCP Server
The external service. A notes app, a calendar, a database — anything that exposes its capabilities through MCP. Hjarni is an MCP server.
Tools
The actions the server offers. Search notes. Create a note. List folders. The AI discovers available tools and decides which ones to use based on your request.
You ask your AI a question. Your AI checks its connected MCP servers. Finds relevant data. Calls the right tool. Uses the result in its response.
You don't manage the connection. You don't write queries. You just talk to your AI.
Who supports MCP
MCP was created by Anthropic (the company behind Claude) and released as an open standard. It's now supported by:
- Claude — Claude.ai, Claude iOS, Claude Desktop, and Claude Code all support MCP
- ChatGPT — supports MCP via the Apps Directory and developer mode
- Cursor, Windsurf, and other coding tools — use MCP for tool integrations
- Perplexity, Gemini, GitHub Copilot, and other assistants keep adding MCP connectors
Any AI that adds MCP support can instantly connect to every MCP server. Including Hjarni. For the canonical list of clients that work with Hjarni, see the full compatibility matrix.
MCP and Hjarni
Hjarni has a built-in MCP server. Not a plugin. Not a third-party integration. The app was built around MCP from day one.
When you connect Hjarni to ChatGPT or Claude, your AI gets access to tools like these:
That is a sample, not the whole set. The MCP server reference documents every tool Hjarni exposes.
Your AI also reads the custom instructions you set on each folder. It doesn't just access your data. It follows your rules.
MCP is the protocol. Hjarni is the memory. Your AI reads your notes instead of starting from scratch.
Common questions
FAQ
Do I need to be technical to use MCP?
No. If you can add a connection in ChatGPT or Claude, you can use MCP. Hjarni handles the protocol. You just talk to your AI.
Is MCP the same as an API?
Similar idea, different audience. An API is for developers. MCP is for AI assistants. Hjarni has both: the REST API for your code, the MCP server for your AI.
Does MCP work with all AI assistants?
Only those that support the protocol, but that is most of them now: ChatGPT, Claude, Perplexity, Gemini, GitHub Copilot, and nearly every AI coding tool. Any AI that adds MCP support works with Hjarni automatically. See the compatibility matrix for the full list.
Is my data safe with MCP?
MCP uses OAuth for authentication. Your AI can only access your notes after you authorize the connection. We don't train AI on your data.
Where can I learn more about the MCP spec?
The official spec is at modelcontextprotocol.io. For Hjarni's implementation details, see the MCP server reference.
Give your AI a memory. Free.
Connect Claude or ChatGPT to notes they can actually read and write.
Give your AI a memory. Free.