# AI Glossary

Short, plain-language answers to the "what is…" questions people ask when they start connecting AI assistants to their own notes. Each entry links to a longer explainer.

## What is MCP?

**Model Context Protocol.** An open standard from Anthropic that lets AI assistants talk to external tools and data through a shared interface. One protocol, many connections: any client that speaks MCP can use any server that speaks MCP.

In practice: you connect Hjarni to Claude or ChatGPT once, and the AI can search your notes, read them, and create new ones from inside the conversation. You don't write code. You just talk.

[Read the full explainer →](https://hjarni.com/docs/what-is-mcp)

## What is a second brain?

**A personal knowledge management system.** A place outside your head where you store ideas, notes, and research so you can find and use them later. The term was popularised by Tiago Forte in *Building a Second Brain*.

In the AI era, a second brain isn't just a personal archive. With MCP, it becomes external memory that your AI assistant can reach into at the start of every conversation.

[Read the full explainer →](https://hjarni.com/docs/what-is-a-second-brain)

## What is RAG?

**Retrieval-Augmented Generation.** A pattern where an AI model fetches relevant documents from an outside source before it writes its answer, so the response is grounded in your data instead of just the model's training set.

A search step happens first. Then the matching snippets are pasted into the prompt as context. Then the model writes. That's it. Most "chat with your documents" products use some form of RAG under the hood.

[Read the full explainer →](https://hjarni.com/docs/what-is-rag)

## What is an AI agent?

**An AI assistant that can decide which tools to call and then call them, in a loop, to finish a task.** Not just "answer my question." More like "go do this and come back when you're done."

Claude Code is an agent. Cursor's agent mode is an agent. A regular ChatGPT chat with no tools is closer to an assistant. The line is whether the model takes actions on your behalf, observes the result, and decides what to do next.

[Read the full explainer →](https://hjarni.com/docs/what-is-an-ai-agent)

## Related terms

A few neighbouring concepts that show up in the same conversations. None get their own page yet, but the short definition is enough for most people.

- **MCP server**: A program that exposes tools and data over the Model Context Protocol so AI clients can call them. Hjarni runs one.
- **MCP client**: The AI side of the conversation: Claude, ChatGPT, Cursor, Claude Code, and so on.
- **Tool use**: The general capability behind agents and MCP: the model can call a function, see the result, and incorporate it into its next response.
- **Context window**: How much text the model can see at once. Large but finite. RAG and MCP are both ways to feed only the relevant slice of your knowledge into it.
- **Embeddings**: A numerical representation of text used for similarity search. The "retrieval" half of RAG usually runs on embeddings.
- **Fine-tuning**: Re-training a model on extra data so it bakes that knowledge into its weights. Slower, more expensive, and harder to update than RAG.
- **LLM**: Large language model. The text-generating model underneath ChatGPT, Claude, Gemini, and the rest.
- **PKM**: Personal knowledge management. The discipline second brain apps belong to.