The second-brain idea has been around for years: capture everything you read, decide, and learn into one system, and stop trusting your memory. The tooling was always the weak point — most second brains become write-only archives, meticulously organized and never opened again. What changed is that now something else can do the reading: an AI agent with direct access to your knowledge.
Why most second brains die
Capture is fun; retrieval is work. A folder of a thousand notes only helps if you remember which note to open, and keyword search fails the moment your query words don't match the note's words. The archive grows, the payoff doesn't, and eventually you stop capturing too. The fix isn't better folders — it's making retrieval someone else's job.
The three pieces
- Knowledge bases hold the durable stuff — documents, articles, video transcripts — converted to clean Markdown and indexed for semantic search, so a question finds the right passage by meaning.
- Memory holds the small stuff — decisions, preferences, facts that came up in conversation. Your agent calls remember to store one and recall to retrieve it later, across sessions.
- MCP is the wiring — it exposes both as tools your agent calls directly, so the same brain works from Claude, Cursor, or any MCP client without per-app glue.
Capture: everything becomes a document
The capture side has to be frictionless or it won't happen. With Kit for AI, anything with text in it goes in through one converter: PDFs and Office files, web pages fetched as clean article text, images via OCR, YouTube videos via their transcripts. Paste a link or drop a file, pick a knowledge base, done. Agents can even capture for themselves — "save this page to my research KB" is one MCP tool call.
Recall: the part your agent does for you
This is where it stops being an archive. Mid-task, your agent searches the knowledge base on its own: ask it to draft a proposal and it pulls your pricing notes, the client call transcript, and that competitor article you saved last month — cited, so you can check its sources. Ask "what did we decide about the auth flow?" and recall surfaces the decision you stored three weeks ago. You never open a folder. How the memory half works under the hood is covered in our guide to cloud persistent memory.
A workflow that sticks
- When you read something worth keeping, capture it — a link takes seconds.
- When you decide something, tell your agent to remember it in one sentence.
- When you work, ask questions instead of hunting files — and let the agent search mid-task.
- Keep separate concerns in separate spaces — research, work, personal — so retrieval stays sharp.
The second brain finally works because you're no longer its librarian — your agent is. Setup is one MCP install; the docs walk through it, and it's free to start.