I Gave an AI Access to My Second Brain

What happens when you point Claude Code at your Obsidian vault? More than you'd expect.

The Link Graveyard

I'm a news junkie. I'm constantly surfing Hacker News, Slashdot (yes, it still exists), Reddit, and lately X - always hunting for interesting technology content. For years, my workflow was the same: find something interesting, text myself the link, and tell myself I'd go through it later. I'd document what I wanted, take proper notes in Obsidian, and eventually catch up.

Except I never caught up. The links piled up. The notes never got written. My "read later" list became a graveyard of good intentions.

Sound familiar?

It wasn't just frustration. It felt like failure. I started questioning the point of saving all this information - articles I wanted to dig into, ideas that might be useful for a project, new ways of doing things. What was the point if it all just sat there in my SMS history? No real way to search for anything. No real way to use it besides scrolling and hoping I'd find what I was looking for.

And here's the kicker: more often than not, I didn't find it. Unless something was from the last few days, it was gone. When I needed that link I'd saved months ago, I'd end up doing a Google search to re-find the information the old-fashioned way. I'd done the work of capturing it, then still had to do the work of finding it again. Double the effort for the same result.

I tried everything. Pocket. Instapaper. Browser bookmarks with careful folder hierarchies. Even a dedicated "Links to Process" note in Obsidian. Nothing stuck. The capture was easy. The processing never happened.

The problem wasn't the tools. The problem was me. I had infinite appetite for collecting and zero bandwidth for organizing.

The First Step: MCP + Obsidian

Then I discovered Claude could connect directly to Obsidian through the MCP (Model Context Protocol) plugin. This was the first breakthrough.

MCP is Anthropic's standard for connecting AI to external tools and data sources. Someone had built a plugin that let Claude read and write to Obsidian vaults. I set it up, gave Claude examples of how I like my notes formatted, and started sending it links instead of myself.

The workflow was simple: I'd find something interesting, paste the link to Claude, and say "capture this." Claude would fetch the content, summarize the key points, format it according to my template, and save it to my Web Captures folder.

Suddenly, Claude was doing the processing I'd been too busy to do myself. My Web Captures folder started filling up with well-formatted notes I actually wanted to read. Each capture had a consistent structure: source info, synopsis, key points, my analysis section, and related links.

The backlog started shrinking. For the first time in years, I was capturing more than I was losing.

But I was getting frustrated with the MCP. While it was better than nothing, it wasn't perfect. It took time to find things, and it could never really pull a lot of information into context. The Claude desktop app kept forgetting I even had the MCP installed unless I reminded it, and then I'd have to re-teach it how I wanted to store information.

The MCP plugin was great for adding new notes, but it couldn't really explore what was already there. It couldn't find connections between notes, reorganize my folders, or synthesize ideas across multiple captures. For that, I needed something with deeper access.

The Leap: Claude Code + Full Vault Access

I'd been using Claude Code for some personal coding projects and had a good idea of how well it worked with larger-scale projects - navigating huge codebases with folders and files everywhere. One day it clicked: if Claude Code can handle massive software projects, why can't it do the same for my Obsidian vault? It's just markdown files in folders.

I could set up a CLAUDE.md file with rules and teach it what it needs to know. I could write skills for handling certain activities. The vault would just be another project.

So I pointed Claude Code directly at my vault. Not through the plugin, but treating the entire Obsidian vault as a project.

That's when everything changed.

Claude Code isn't just a coding assistant - it's AI with full filesystem access. When you point it at a folder, it can read everything, search everything, create files, move files, and understand the structure of what's there. The "code" in the name is underselling it. It's really "Claude with access to your filesystem."

My vault has over 600 files. Web Captures from years of link-hoarding. Project ideas in various states of abandonment. Work notes. Even personal journals going back to 2001. I added a CLAUDE.md file with some basic context and let Claude loose.

The first thing I had it do wasn't capturing new links. It was reorganizing the chaos I already had.

My vault was a mess - folders full of folders full of other folders and files. It was hard to tell what was actionable versus just stored information. Claude actually suggested the PARA system, set everything up, and we went through it together, organizing things as we went.

It was night and day. Suddenly, if I needed something on a project I was working on, I knew exactly where it would be. If I needed information I'd captured about a local restaurant, it was under my Resources folder. Things were actionable. I could find things - and with Claude's help, track them down even faster.

Obsidian went from something I used from time to time to an almost constantly-used tool in my inventory.

It's not perfect. But it's been one of the most life-changing things I've done with AI.

Quick Aside: What's PARA?

If you're not familiar with PARA, it's an organizational system created by Tiago Forte. The acronym stands for:

• Projects: Active work with specific outcomes and deadlines
• Areas: Ongoing responsibilities you maintain over time (health, finances, work)
• Resources: Reference materials on topics you're interested in
• Archive: Completed or inactive items from the other three categories

The key insight is organizing by actionability, not by topic. A note about "machine learning" might go in Projects (if you're actively building something), Areas (if it's part of your ongoing work), Resources (if it's reference material), or Archive (if it's from a completed project).

My folder structure looks like this:

MainVault/
├── 01 - Projects/      # Active projects with specific outcomes
├── 02 - Areas/         # Ongoing responsibilities (Work, Personal)
├── 03 - Resources/     # Reference materials, books, web captures
├── 04 - Archive/       # Completed/inactive items
├── Inbox/              # Quick capture, unsorted items
├── Journals/           # Daily notes, personal journals
└── Templates/          # Note templates

When Claude understands this structure, it can make intelligent decisions about where things should go. "File this project idea" doesn't need me to specify the path - Claude knows it belongs in 01 - Projects/.

I Don't Really Use Obsidian Anymore

Here's something I didn't expect: working with Claude Code and Obsidian directly changed everything - including how I access my vault.

I don't use the Obsidian app nearly as much as I used to. Now, I typically work with my vault using VS Code. It may not be as pretty and refined, but everything is still there at my fingertips - and Claude Code is right there with me.

I still use Obsidian on my phone for quick captures. I'll take a note in the inbox, then revisit it with Claude once I get to my laptop. Claude helps me organize it or add additional information as needed.

So the workflow became: phone for capture, Claude Code for processing. Obsidian is still the storage layer, but Claude Code became the interface I actually use.

What It Can Actually Do

Organize

"Hey Claude, I want to start a blog at lutherlowry.com. Can you set up a folder structure for posts?"

Done. Folder created at 02 - Areas/Personal/lutherlowry.com/Posts/, nested in the right place in my PARA hierarchy. When I finish a draft, I say "file this away" and it moves to the right location with a proper date-stamped filename.

No more "I'll organize this later." The organization happens as part of the conversation.

This extends to cleanup tasks too. "My inbox has 47 items in it. Help me triage." Claude reads through them, suggests where each should go, and moves them when I approve. What used to be a dreaded monthly chore becomes a quick conversation.

Find

"What have I captured about AI security?"

Claude searches across 600+ files and surfaces relevant notes I'd completely forgotten about. Not just keyword matching - it understands what I'm looking for and finds related content even when the exact words don't match.

This is the feature that changed how I think about my vault. It went from a write-only archive - a place where information went to die - to something I actually use. The knowledge I've accumulated over years is suddenly accessible.

"Have I written about this before?" Finally has a real answer.

Synthesize

This is where it gets powerful.

"I want to write a blog post. Find something interesting in my Web Captures."

Claude scanned my captures, surfaced five options with summaries, and when I picked one, we built an outline together from the key points. I wrote the draft, Claude helped me research statistics to back up the claims, found academic sources, and helped integrate them with proper citations.

A single Reddit thread I'd saved became a researched, sourced blog post. The raw material was already in my vault - Claude just knew how to find it and transform it.

This is the synthesis I never had time to do myself. All those captured links, all those "this is interesting, I should do something with this" moments - Claude can actually follow through on them.

Think With Me

"Hey Claude, I have an idea for a project."

We talk through it together. What is it? Who's it for? What's interesting about it? What are the risks? What would an MVP look like? What similar things have I captured or thought about before?

Once the idea is fleshed out, Claude creates a note and files it in the right PARA folder.

The conversation IS the note-taking process. There's no separate step where I have to go write things down. The thinking and the capturing happen together. And because Claude can search my vault, it can bring in relevant context from things I've forgotten I knew.

Interview Me

This one surprised me.

I've been working on an autobiography project - capturing memories, stories, and reflections before they fade. The problem is, I don't know what I remember until someone asks the right question.

Claude became that someone.

"Tell me about your friend groups growing up." I start talking. Claude asks follow-up questions - who was in the group? What did you do together? What happened to them? It digs into threads I mention in passing. It notices when I skip over something and circles back.

It's like having a biographer on call. A reporter who's genuinely curious about your story and knows how to draw it out. When we're done, Claude organizes everything into structured notes - timelines, character sketches, thematic threads.

The memories were always there. I just needed someone to ask.

Capture With Context

Here's something that changed how I save information: Claude doesn't just summarize what I send it. It asks me questions.

Why did I find this interesting? How might I use this? Does it connect to anything I'm working on?

My captures now include my opinions and thoughts alongside the summary. Those opinions are some of the best bits when I go back later - they tell me why I saved something, how I planned to use it. It's a nice reminder of my intent, which was exactly what got lost in the SMS graveyard.

I'm saving a lot more than ever now. Claude automates the busywork of formatting and filing, but it also enriches each capture by prompting me to think about what I'm saving. My future self gets my perspective, not just the source material.

A Real Example: Today's Session

I'm writing this post with Claude's help. Here's what actually happened in this session:

1. "Find something interesting in my Web Captures for a blog post" - Claude searched my vault and surfaced five technical topics with summaries of each
2. "Let's do the vibe coding security one" - We worked together to draft an outline based on the source material
3. I filled in the details - Writing the sections, adding my perspective, expanding the points
4. "Can you fact-check these statistics?" - Claude searched the web, found academic sources (USENIX, CodeRabbit research), and helped me integrate proper citations
5. Claude edited and refined - Smoothing out rough spots, suggesting improvements, catching inconsistencies
6. "File it away under lutherlowry.com" - Claude created the folder structure and moved the file to the right place
7. "Let's plan the next article" - Which became this post you're reading

Zero lines of code written. All knowledge work. All inside my Obsidian vault.

Why This Is Different from ChatGPT

I use ChatGPT too. It's great. But for knowledge management, the difference is substantial:

The fundamental difference: with ChatGPT, you're always starting from zero. You paste in context, explain your situation, upload relevant files. It's like talking to someone with amnesia who's very smart but knows nothing about you.

With Claude Code pointed at your vault, you're talking to something that already knows your projects, your captures, your ideas, your organizational system. It can build on what you've already done instead of starting from scratch.

Your second brain stops being a static archive. It becomes a conversation partner.

What Doesn't Work (Honest Assessment)

It's not all magic. Here's where things break down:

Context limits are real. Claude can't hold your entire 600-file vault in memory at once. It searches, reads relevant files, and works with what it finds. Sometimes it misses things that would be relevant because they didn't match the search.

It gets the PARA system wrong sometimes. "Is this a Project or an Area?" is a judgment call that Claude doesn't always get right. I still need to review where things get filed.

Complex reorganization is tricky. "Reorganize my entire Resources folder" is too big. You need to break it into smaller tasks and guide the process.

It can be confidently wrong. Claude might tell you "I found 3 notes about X" when there are actually 7. Always verify when it matters.

It's not a replacement for thinking. Claude can find, organize, and synthesize - but the ideas still have to come from you. The judgment about what matters still has to come from you.

The key is calibrating your expectations. It's a very capable assistant, not an autonomous agent. You're still driving.

Before You Do This: The Caveats

I'm not telling you to go do this tomorrow.

Claude Code has full filesystem access. That's what makes it powerful, but it's also what makes it risky. It can create files, move files, delete files, and modify anything in your vault. Things can go wrong. A misunderstood instruction could move files to the wrong place, overwrite something important, or delete notes you meant to keep. You're giving an AI write access to your knowledge base - treat that with appropriate caution.

We don't fully know what happens to the data Claude sees. Anthropic has privacy policies, but policies change. AI is a new technology, and who knows what could happen to this data in the future. There's always a chance your information could be used in ways you didn't expect, or that a breach exposes something you'd rather keep private.

I made a conscious choice. My vault doesn't contain anything extremely personal. No passwords, no financial details, nothing that would be life-altering if it somehow got out into the world. My Web Captures are mostly public articles I found interesting. My project notes are ideas, not secrets. My old journals are... well, embarrassing, but not dangerous. I didn't do any special "cleanup" before pointing Claude at my vault. I was already comfortable with what was there.

I feel comfortable that my data is pretty safe and secure. But not everyone feels the same, and while I'm comfortable, I don't want to recommend people do exactly what I did. I'm responsible for my data in my vault, but I don't want to be responsible if something goes wrong with someone else's.

If your vault contains:

• Medical records
• Financial information
• Client or work confidential data
• Truly private journal entries
• Credentials or API keys

...then think hard before pointing an AI at it. Maybe create a separate vault for the stuff you're comfortable sharing. Keep sensitive information somewhere Claude can't see.

If you want to try it? Go for it. Just know that once you do this, you're not in full control of your data anymore. At the very minimum, make sure you've got a backup.

If You Want to Try It

The Basic Setup

1. Install Claude Code (Anthropic's CLI tool)
2. Navigate to your Obsidian vault folder in the terminal
3. Run claude to start a session with your vault as the working directory
4. Claude now has access to everything in that folder

That's it. No plugins required for this approach - Claude Code works directly with the filesystem.

The CLAUDE.md File

Create a file called CLAUDE.md in your vault root. This gives Claude persistent context about your setup. Here's a simplified version of mine:

# CLAUDE.md - Obsidian Vault Context

## Vault Overview
This is a personal Obsidian vault organized using the PARA system.

## Directory Structure
- 01 - Projects/    # Active projects with specific outcomes
- 02 - Areas/       # Ongoing responsibilities
- 03 - Resources/   # Reference materials, web captures
- 04 - Archive/     # Completed/inactive items
- Inbox/            # Quick capture, unsorted items
- Journals/         # Daily notes, personal journals

## Conventions
- Use date-prefixed filenames: YYYY-MM-DD - Title.md
- Web captures go in 03 - Resources/Web Captures/YYYY/MM
- Respect the PARA hierarchy when filing new notes

Claude reads this file and understands your vault's organization. The more context you provide, the better it can work within your system.

Good Starting Prompts

• "What's in this vault? Give me an overview."
• "Find all notes related to topic"
• "I have an idea for a project - let's talk through it and then file it away"
• "Summarize what I've captured about subject"
• "Help me triage my inbox folder"
• "What projects have I abandoned that might be worth revisiting?"
• "Find connections between note A and note B"

The Bigger Picture

Claude Code is marketed as a coding tool. And it's excellent at that. But the real unlock is simpler: it's AI with filesystem access.

That means anything you can organize in folders and files becomes accessible to AI. Your notes. Your research. Your ideas. Your second brain.

More and more, my Obsidian-enabled Claude Code is almost like having a conversation with myself. We grok out ideas together. We discuss the things I've found and how they might connect to other items in my vault. When I have an idea, I ask if there's anything in my vault that might make it even better. How can we combine ideas?

Especially in my story ideas folder - there's so much in there. While I might have an interesting tale I want to spin, combining it with another idea I had weeks ago sometimes makes it a hundred times stronger. Claude can see those connections when I miss them.

That's what this really is: a second brain I can actually talk to.

The constraint isn't the tool anymore. It's our imagination about what to point it at.

For a news junkie with a link graveyard problem, that's been transformative. The links still pile up - I haven't cured my content addiction. But now they actually turn into something. The friction between "this is interesting" and "this is useful" has dropped dramatically.

That's what happens when you give an AI access to your second brain.

What would you do if AI could access your notes? I'm curious what use cases I haven't thought of yet.