Workflow & Tools

Claude + Obsidian: The Second Brain Setup That Actually Gets Used

Published May 8, 2026 · 11 min read

I've watched a lot of people set up Obsidian as their "second brain." I've watched almost none of them use it past month two.

The pattern is always the same. Week one is Plugins Week — the new user installs Dataview, Templater, Excalidraw, the Kanban plugin, the Calendar plugin, and three other things that all promise to be transformative. Week two is Folder Architecture Week — twelve nested folders, a tagging system with fourteen levels, a PARA implementation, a Zettelkasten daily ritual. Week three is Capture Week — they're highlighting in browsers, clipping articles, sending voice memos. Week four they have 280 notes. By month three they have 600 notes and zero retrieval. By month four they're back to using Notion.

The diagnosis is always the same too: they built a capture system, not a retrieval system. They optimized the part where stuff comes in. They never solved the part where stuff comes back out — which is the only part that actually changes how you think.

Claude solves this. Not with a plugin. Not with a community-vault. With one decision and three prompts. Here's the setup I actually use, after eight months of an Obsidian vault that gets opened every day.

The single insight that fixes 90% of second brains

Stop trying to organize the notes for retrieval. Let Claude be the retrieval.

Every Obsidian tutorial you've watched has been arguing about folder structures, tagging systems, and linking strategies. They're arguing about it because human-only retrieval — actually finding the note you wrote four months ago — depends entirely on how you filed it. Bad filing, bad retrieval. Good filing, good retrieval. So the tutorials obsess over filing.

Claude doesn't care how you filed it. Claude can read your entire vault and tell you what's in it. The whole "how do I structure this?" debate disappears the moment you stop being the search engine.

That's the unlock. Once you accept it, you can throw away most of the architecture overhead, capture more carelessly, and still get more out of your notes than you did when you spent forty minutes a week curating the structure.

The setup, end to end

Step 1 — Collapse your folder structure to four folders

Not twenty. Not seven. Four.

• Daily/ — one note per day, append-only. Anything you capture during that day goes here. No exceptions, no sub-categorizing.
• Projects/ — one folder per active project. Active means you've touched it in the last two weeks.
• Reference/ — anything you'll re-read more than once: SOPs, checklists, frameworks, voice docs.
• Archive/ — projects you've closed. Move them here when they're done. Don't delete.

That's the entire structure. Tagging is optional and almost always unnecessary. Linking is optional and helps when it's organic, but is no longer load-bearing — Claude will find related notes whether you linked them or not.

This shift alone takes about ninety percent of the friction out of capture, because you no longer have to decide where anything goes. You just write it in today's daily note and move on.

Step 2 — Make the daily note your only capture surface

Open Obsidian. Open today's daily note. That's where everything goes.

• A meeting note → write it as a section in today's daily note
• An idea you had in the shower → bullet under today
• A quote from a book → block quote under today
• A decision you made on a project → noted under today
• A research find → linked or summarized under today

If something graduates into something bigger — a project, a SOP, a piece you want to publish — then you create a new note in Projects/ or Reference/ and pull the relevant bits over. But that's a deliberate, second-pass move. The first pass is always the daily note.

This single rule is the most important behavioral change in the entire setup. Most second brains die from filing fatigue. Removing the filing decision from the capture moment removes the fatigue.

Step 3 — Connect Claude to the vault

You have three options here, in order of friction:

1. Drag-and-drop into Claude.ai (lowest friction). Open the day's relevant notes in Finder, drag them into the Claude chat. Works for ad-hoc questions. Slightly tedious for everything-vault queries.
2. Claude Projects with the vault as the knowledge base (best for daily use). Create a Project called "Vault." Upload the most-referenced reference docs (your voice doc, your SOPs, your project briefs) once. Then drag the daily note for any session-specific question. Setup time: 15 minutes. This is where most people end up.
3. API + script (highest leverage, takes a developer). A small script that watches the vault and feeds new notes into a Claude API call with prompt caching. Cheap, fast, runs in the background. Worth it if you have a vault past 500 notes and you're a developer. Skip if not.

For 90% of people the right answer is option two. The Project becomes the lens. The vault stays in Obsidian. Claude becomes the retrieval layer.

Step 4 — Use the three retrieval prompts that do all the work

These are the only three prompts I run against my vault. Every other use is a variation.

The "find it" prompt:

I'm looking for a note from somewhere in the last 6 months
about [topic]. I remember it had something to do with [detail].
Quote the relevant passages back to me with the dates of the
notes they came from.

This single prompt replaces 90% of the manual searching I used to do. Claude finds things I would never have surfaced because the relevant note was in a daily note from a Tuesday in March that I'd long forgotten.

The "summarize this" prompt:

Read everything I've written about [project / topic] in the
last 30 days. Summarize where I am, what's blocked, what
decisions I've made, and what I said I'd do next. Then ask
me one clarifying question.

This is the Monday morning prompt. Five minutes of Claude reading your project notes and you start the week with a clearer head than two coffees and a planning ritual.

The "connect this" prompt:

Across all my notes, find 3-5 ideas, decisions, or quotes
that connect to what I'm working on right now: [paste current
context]. Surface things I haven't explicitly linked yet.

This is the prompt that turns the second brain from a filing cabinet into something generative. Claude finds connections you weren't tracking. About one in three runs surfaces something genuinely surprising — a quote from two months ago that reframes the project you're working on today.

Three prompts. That's the entire usage layer. Master these and you'll never go back to manual searching.

The five things you stop doing

Once Claude is the retrieval layer, several rituals you used to defend become wastes of time. Drop them.

1. Manual tagging. Tags were a hack to make notes findable. They're now redundant. The only tag worth keeping is one that marks something as a "live project" vs. "archive" — and even that is mostly a folder concern.
2. Note-to-note linking as a capture ritual. Organic links are still good. Forced "you must link every new note to three existing notes" rituals (popular in Zettelkasten communities) are dead. Claude finds connections without your bookkeeping.
3. The MOC (Map of Content) habit. The "make a top-level overview note for each domain" practice was great when humans were the retrieval engine. It's now just maintenance overhead. Skip.
4. Plugins that organize. Dataview, Templater, Breadcrumbs — most of them existed to compensate for the retrieval problem. With Claude doing retrieval, you can run a vault on stock Obsidian and never feel the lack.
5. Weekly reviews focused on filing. Replace with a five-minute Claude session — "summarize this week, surface what I missed" — and you get more insight than an hour of folder shuffling.

What stays — and gets more important

Two habits are more important once Claude is doing the retrieval, not less.

First: capture more, not less. The whole reason note-takers used to under-capture was the filing tax — every note had a setup cost. With the daily-note-only rule, the cost is zero. Capture everything: shower thoughts, half-formed ideas, screenshot notes, conversations, questions. The vault gets denser, and Claude gets more to work with. The retrieval gets better the more you write.

Second: write in your own voice. Notes that sound like you (sentence fragments, informal, jokes-to-self) retrieve better than notes that sound like a textbook. Claude can't surface something you wrote like a corporate memo, because every note from your vault sounds the same. Notes that sound like you create distinct hooks Claude can latch onto.

What this actually feels like, after eight months

I open Obsidian once a day to dump into the daily note. I open Claude when I want to find, summarize, or connect. The two systems coexist with almost no overlap, and zero ritual.

Concretely, what's different:

• I find old notes in 30 seconds instead of 6 minutes
• I never lose an idea because I had to decide where to file it
• Monday-morning project re-loading takes one prompt instead of forty minutes of tab-shuffling
• I find connections I would never have surfaced manually — at least one a week, sometimes more
• The vault has gotten denser, not messier, because density is now an asset rather than a liability

The last point is the one that took me longest to internalize. In the old model, more notes meant more chaos. In this model, more notes means a smarter retrieval layer. The incentive on capture inverts — and the second brain finally starts compounding the way the books promised it would.

The unpopular take

If you're someone who has spent the last two years tweaking your Obsidian setup — building MOCs, refining tag taxonomies, learning Dataview queries, watching YouTube videos about PARA vs. ACCESS vs. Zettelkasten — most of that work is now obsolete. That's the part nobody in the second-brain community wants to say out loud, because it invalidates a lot of identity work.

The part nobody's saying out loud is what changed: the retrieval bottleneck moved. It used to live with the human, which meant filing rituals were load-bearing. It moved to the model, which means filing rituals are decorative. The setups that survive this shift are the ones that minimize the human's filing cost — because the model doesn't care anyway.

Throw out the architecture. Keep the daily note. Let Claude be the index. That's the entire setup.