Agentic PKM: Claude + Obsidian, From Notes to Published Work

How do you connect Claude to Obsidian? Three ways, depending on the tool: open Claude Code (the terminal agent) in your vault folder and it reads/writes your actual .md files directly — no plugin needed. Point Claude Desktop at your vault with the official MCP filesystem server, and it gets the same file-level access through a chat window. Or install the Obsidian MCP plugin (obsidian-claude-code-mcp) so either tool goes through Obsidian’s own runtime — links, templates, and sync stay intact. All three are covered below.

Agentic PKM is personal knowledge management with an AI agent working directly inside your knowledge base. Instead of copy-pasting notes into a chat window, an agent like Claude Code opens your Obsidian vault as its working directory — it reads your actual markdown files, knows your voice and your frameworks from a briefing file, and creates finished outputs (scripts, posts, newsletters, diagrams) right back into your vault. The point isn’t smarter note-taking. It’s activation: closing the gap between “I have an idea” and “I just published something,” so the knowledge you’ve spent years capturing finally becomes visible, published work.

This is the complete playbook for the 9-to-5 Architect who has a vault full of knowledge and nothing published to show for it — and is ready to fix that permanently using Claude + Obsidian.

The Diagnosis

You Don’t Have a Knowledge Problem. You Have a Visibility Problem.

Your vault is a cemetery of almost-ideas. Every note you’ve ever taken is a thought that almost became something. A book highlight that almost became a framework. A shower thought that almost became a video. A client insight that almost became a course.

The PKM industry has sold you the idea that better organization is the answer. More folders. Better tags. A perfect MOC structure. But organization isn’t your bottleneck. Activation is.

Activation is the gap between having an idea and doing something with it — between capturing a thought and turning it into a YouTube video, a LinkedIn post, a newsletter, a paying offer. Most PKM systems stop at capture. None of them close the activation gap.

Claude + Obsidian, set up the right way, closes that gap. Not by replacing your thinking — but by handling everything between “I have an idea” and “I just published something.”

The vault stores what you know. Claude deploys it. The combination turns a cemetery of almost-ideas into a publishing machine.

The Progression

The Three Levels: You Don’t Start at the Deep End

The Agentic PKM system has three levels. Most people try to skip to Level 3 and burn out. The right move is to build each level completely before advancing. Each one delivers immediate wins — and makes the next level dramatically more powerful.

1. 01
   Level 01 — The Thinking Partner. Use Claude.ai as a smart collaborator. Paste notes, ask questions, get help developing ideas. No setup required beyond an account. Time to first win: Day 1. What you unlock: context-aware answers, voice learning, draft generation, idea development.
2. 02
   Level 02 — The Vault AI System. Install three components in Obsidian: a CLAUDE.md brain briefing, an +Inbox/ capture folder, and a _prompts/ library. Claude now starts every session knowing who you are. Setup: 45 minutes. First win: same session. What you unlock: permanent context, repeatable workflows, voice-locked outputs, script templates.
3. 03
   Level 03 — Agentic PKM. Install Claude Code. It reads your actual vault files, creates outputs directly in Obsidian, builds Excalidraw diagrams, and runs multi-step workflows without managing every prompt. Setup: 1 hour. What changes: publishing becomes inevitable. What you unlock: direct vault access, auto-file creation, Excalidraw builds, content multiplication.

Level 01

Level 01 — The Thinking Partner

Most people use Claude like a vending machine: insert question, receive answer, walk away. This is why they get generic outputs. They gave Claude nothing — no context, no voice, no knowledge of who they are or what they’re building — and they got nothing specific in return.

The PKM person’s advantage is profound: you already have years of thinking stored in structured notes. That vault is exactly the context Claude needs to stop being generic and start being genuinely useful to you specifically.

At Level 01, you do this manually. You paste your notes, explain your audience, share samples of your best writing. The results are immediately better than what most people get from AI — because you’re giving Claude something to work with.

Try This — Level 01 (No Setup Needed)

I'm going to give you three samples of my best writing. Study them carefully — the rhythm, the word choices, the sentence length, the tone, the things I never say. Then confirm what you've learned about my voice by describing it back to me in 5 specific observations. [Paste Sample 1] — [Paste Sample 2] — [Paste Sample 3]

What comes back: a precise description of your voice — things you’d never been able to articulate yourself. From this point on, every output in this conversation sounds like you.

Level 02

Level 02 — The Vault AI System

Level 02 removes the copy-paste friction. Instead of re-explaining yourself every session, you write your brain briefing once and Claude reads it automatically. Instead of starting from scratch with every prompt, you pull from a library of tested templates for your most common workflows.

The CLAUDE.md Brain Briefing

A CLAUDE.md file lives at the root of your Obsidian vault. It’s a plain Markdown document — one to two pages — that tells Claude everything it needs to know before you say a word. Your voice. Your audience. Your frameworks. Your goals. What you’re building and why.

Claude Code reads this file automatically when it opens your vault. Claude.ai can read it if you share the file at the start of a conversation. Either way: no more re-explaining yourself. No more generic outputs.

What to include in your CLAUDE.md:

• Who I am: Your name, role, credentials, background
• My audience: Who reads/watches your work and what they struggle with
• My voice: 5 specific observations about how you write and speak
• My frameworks: The named ideas and models you teach
• My platforms: YouTube, LinkedIn, newsletter — what you publish where
• Current focus: What you’re building right now

The +Inbox/ and _prompts/ Structure

Two folders complete the Level 02 system. +Inbox/ is where everything new lands — voice memo transcripts, shower thoughts, quick captures, highlights. It’s the funnel. You don’t process here, you just preserve.

_prompts/ holds your template files — one Markdown file per workflow. “Script from inbox note.” “Multiply this script into a content week.” “Name the framework hiding in these notes.” These templates carry all the instructions so you don’t have to re-type them.

The Vault AI System doesn't change what Claude is capable of. It changes how much Claude already knows about you before the conversation starts. That's the entire difference between generic and genuinely useful.

Level 03

Level 03 — Agentic PKM

Level 03 is where publishing stops feeling heroic and starts feeling inevitable. Claude Code opens your Obsidian vault as its working directory. It reads your notes directly — no copy-paste. It creates files and saves outputs to the right folders automatically. You describe what you need. The system builds it.

This is not a replacement for your thinking. It’s a replacement for the activation friction that stops your thinking from becoming visible work.

The System

The Mechanics

How to Connect Claude to Obsidian (3 Ways)

Everything above describes what becomes possible. Here’s how the connection actually works, mechanically, for each tool.

Way 1 — Claude Code, opened in your vault folder

The simplest connection: no plugin, no config file. Open a terminal, cd into your Obsidian vault, and run claude. Claude Code treats the vault as its working directory — it can read, search, and write your actual .md files directly, the same way it reads and edits code in a repo. This is what Level 02 and Level 03 above are built on.

Way 2 — Claude Desktop, via the MCP filesystem server

Claude Desktop is a chat app, not a terminal agent, but it can reach local files through MCP (Model Context Protocol). Anthropic’s official filesystem server points Claude Desktop at one folder — your vault — and nothing else:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/your/vault"]
    }
  }
}

Restart Claude Desktop and it can read and edit notes in that folder — plain files, the same way Way 1 does, just from a chat window instead of a terminal.

Way 3 — The Obsidian MCP plugin (works with either)

Ways 1 and 2 touch files directly on disk. The obsidian-claude-code-mcp community plugin instead runs an MCP server inside the running Obsidian app, so Claude connects through Obsidian’s own runtime — moves preserve links, creates apply templates, and it reflects your live sync/plugin state. Full setup steps, including the Claude Desktop config and the Claude Code /ide command, are in the Obsidian CLI guide.

Pick One

Claude Desktop vs Claude Code for Obsidian

If you’re doing Level 01 or asking occasional questions, Claude Desktop is enough. If you want the Level 02/03 system in this guide — automatic context, saved outputs, multi-step pipelines — Claude Code is the tool built for it.

Do This First

Give Claude Access to Your Vault Safely

Full read/write access to years of notes is a lot of trust. A few guardrails before you connect anything:

• Scope the connection to the vault, not your whole drive. The filesystem MCP config above takes one folder path — use your vault path, not your home directory.
• Keep your vault under version control or Obsidian Sync. A git init in your vault (or Obsidian’s built-in file history — see obsidian history in the Obsidian CLI guide) means any bad edit is one command away from undone.
• List what’s off-limits in your CLAUDE.md. A short “Do not edit” section — Finances, legal, client folders — is often enough; Claude Code reads and respects it like any other instruction.
• Read what it’s about to change before you approve it, especially early on. Claude Code shows you the diff before writing; don’t rubber-stamp it out of habit.
• Start read-only. Ask Claude to search and summarize for a session or two before you let it create or edit files. Trust the connection before you trust it with write access.

Putting It Together

A Claude + Obsidian Workflow

Here’s what the connection looks like in practice, once it’s live. Say you’ve got a rough voice-memo transcript sitting in +Inbox/:

$ claude
# Claude reads CLAUDE.md automatically — your voice, audience, and frameworks
$ claude “Read the newest note in +Inbox/. Turn it into a YouTube script using
my template, and save it to Content/Scripts/.”
Reading +Inbox/2026-08-27 — voice memo.md…
Drafting script using YouTube Script template…
✓ Saved: Content/Scripts/Script — [topic].md

That’s the whole loop: capture, connect, activate. No copy-paste between apps, no re-explaining your voice. For the complete build — CLAUDE.md, folder structure, and the full walkthrough end to end — see Second Brain: Obsidian + Claude Code.

The Full Pipeline: Every Stage, Every Tool

Here’s the complete Agentic PKM pipeline when all three levels are running. Every stage feeds the next. The whole system runs on ideas you’re already having — you’re just not losing them anymore.

1. Capture — Anywhere · Voice Note / Handwriting / Quick Text

Idea captured raw. Shower thought. Book reaction. Client observation. Recorded in 60 seconds, transcribed automatically, lands in +Inbox/. Zero friction at capture — you’re preserving, not developing.

2. Store — Obsidian · +Inbox/ Folder

Raw note sits in the vault. The note lives in +Inbox/ until you’re ready. No action required. Your CLAUDE.md and prompt library are already in place — the system is waiting, not you.

3. Activate — Claude Code · Reads CLAUDE.md Automatically

Claude activates with full context. You open Claude Code. It reads your brain briefing. It knows who you are, what you’re building, what your voice sounds like. You say “process the inbox.” It reads the note, runs the script template, produces the draft — in your voice, saved to Scripts/.

4. Build — Claude Code · Excalidraw + Visual Assets

Framework becomes a visual. If the script contains a named framework or process, Claude Code builds the diagram. One command. Your brand colors, labeled nodes, clean flow. Opens natively in Obsidian with the Excalidraw plugin. Ready to share.

5. Multiply — Claude Code · Content Multiplier Prompt

One script becomes a full content week. The Content Multiplier runs on the script. Five pieces of content saved to Posts/ — three LinkedIn posts, one email newsletter, one short-form. Your editorial calendar fills itself. You schedule and publish.

The Playbook

The Five Core Prompts: Copy These, Run Them Today

These are not “example prompts.” These are the exact prompts that produce real results for PKM people who’ve implemented this system. Copy them into your _prompts/ folder. Run them on your actual vault notes.

The Brain Dump Transformer — Level 01 · No Setup Needed

You have a raw, messy note — half-formed thoughts, incomplete sentences, ideas that felt important when you wrote them but now look like noise. This prompt extracts the signal.

Here's a raw note I captured: [paste note]. Don't rewrite it or clean it up. Just do three things: (1) identify the single most interesting idea buried in here — the one with the most potential — and state it in one sharp sentence, (2) tell me the one question that would unlock this idea further, (3) suggest the best content format for this idea: video, LinkedIn post, newsletter, or framework doc. Give me just those three things. Nothing else.

The Voice Calibration — Level 01 · Run This First, Once

The single most important prompt you’ll ever run. Do this before anything else. It teaches Claude your voice and every output from this point forward gets dramatically better.

I'm going to give you three samples of my best writing. Study them carefully — the rhythm, the word choices, the sentence length, the tone, the things I never say. Then confirm what you've learned about my voice by describing it back to me in 5 specific observations. Here are the three samples: [paste Sample 1] — [paste Sample 2] — [paste Sample 3]

The Framework Hunter — Level 02 · Requires Vault Notes

You’ve been thinking about the same thing for months across different notes and you know it’s connected but you can’t see the shape of it. This prompt finds the invisible framework.

Here are [3–5] notes from my vault that feel connected but I can't articulate why: [paste notes separated by ---]. Do not rewrite them. I want you to: (1) identify the single underlying idea that connects all of these — state it in one sentence as if naming a framework, (2) give that framework three possible names — each under five words, distinct from each other, (3) write a 100-word description of the framework using my voice, (4) tell me what the framework is still missing — what example, evidence, or counterargument would make it complete.

The Inbox-to-Script Pipeline — Level 02 · Your Weekly Workhorse

The prompt you run every time a raw note needs to become a video. Works on voice transcripts, shower thought notes, rough brain dumps — anything in your Inbox.

Using everything you know about my voice and audience from our context brief, take this raw note and turn it into a YouTube script: [paste note or filename]. The script needs: a hook that creates real tension in the first 10 seconds without starting with "In this video" or "Today I want to", three clearly named sections each building on the last, one concrete example or walkthrough in section two that makes the concept tangible, and a CTA at the end directing people to [your CTA]. Write it as spoken dialogue — contractions, short sentences, the word "you." No bullet points. No headers in the script itself. Target length: 600–900 words.

The Content Multiplier — Level 03 · Full Agentic PKM

The most powerful prompt in the system. One video script becomes an entire week of platform content — all in your voice, all saved to the right folders, all ready to schedule. Run this after every video.

Read the script at [filepath or paste]. From this single script, create all of the following and save each to the correct folder in my vault: (1) LinkedIn Post A — a 200-word teaching post that develops the core insight, opens with a bold claim, ends with a reflection question, no bullet points, (2) LinkedIn Post B — a framework post showing the process as 3–4 numbered steps, each one sentence, (3) LinkedIn Post C — a 100-word contrarian take that challenges a common belief in my space, (4) one email newsletter 300 words maximum, opens with a scene not "this week I want to share", includes the insight + one real example + a specific CTA, (5) one short-form post under 60 words that's quotable and sharp. None of them should start with the word "I". None should use: journey, hustle, grind, game-changer, level up.

The Transformation

Before and After: Your Life With the System

This is what shifts. Not just in your workflow — in your identity. The 9-to-5 Architect stops being someone with a full vault and an empty output, and becomes someone whose knowledge is visibly, undeniably at work in the world.

"The identity shift isn't 'I use AI now.' It's 'My knowledge is finally visible — and it's paying off.'"

Common Questions

FAQ

What is Obsidian Claude PKM?

It’s shorthand for this whole setup — using Claude (Claude Code, Claude Desktop, or Claude.ai) as an active agent inside an Obsidian-based PKM system, instead of a chat window you paste notes into. The three levels above are the on-ramp: Thinking Partner → Vault AI System → Agentic PKM.

How does Claude Obsidian integration actually work?

At the file level. Claude Code opens your vault folder directly and reads/writes the .md files inside it. Claude Desktop reaches the same files through an MCP server. The Obsidian MCP plugin goes one layer deeper and connects through Obsidian’s own runtime. See How to Connect Claude to Obsidian above for all three.

How do I connect Claude to Obsidian?

Fastest path: open a terminal, cd into your vault folder, and run claude — Claude Code now reads and writes your notes directly. For Claude Desktop, add the MCP filesystem server to claude_desktop_config.json and point it at your vault path. Full steps and config are in How to Connect Claude to Obsidian above.

What’s the difference between Claude and Obsidian on their own?

Obsidian stores your notes; it doesn’t write anything for you. Claude can write and reason, but starts every conversation with zero memory of your vault unless you give it access. Connected together — CLAUDE.md briefing, vault access, prompt library — each one covers the other’s gap: Obsidian supplies the knowledge, Claude supplies the activation.

What does a Claude Obsidian workflow look like day to day?

Capture stays manual and fast (voice memo, quick note, highlight — lands in +Inbox/). Everything after that runs through Claude Code: process the inbox note into a draft, build the diagram, multiply the script into a week of posts, save it all to the right folders. See A Claude + Obsidian Workflow above for the exact commands.

How do I give Claude access to my vault safely?

Scope any MCP connection to your vault folder specifically, not your whole drive. Put your vault under git or lean on Obsidian’s own file history. Keep a “do not edit” list in your CLAUDE.md for sensitive folders. Start read-only for the first session or two before you let Claude create or edit files. Full guardrails in Give Claude Access to Your Vault Safely above.

Does Claude Desktop work with Obsidian?

Yes — via MCP. Add the official filesystem server (or the obsidian-claude-code-mcp plugin for runtime-level access) to claude_desktop_config.json, restart Claude Desktop, and it can read and edit notes in your vault from the chat window. It won’t read your CLAUDE.md automatically the way Claude Code does — you’d share it manually or rely on the vault files Claude fetches.

Pick Your Path

Work With Frank: One Transformation, Three Levels of Support

Everything in this guide is buildable today on your own. These are the ways to go faster, with more structure, or with Frank directly — depending on where you are and how fast you want to move.

The Creator Vault — Free Forever

Pre-built Obsidian vault, Notion capture dashboard, and AI synthesis prompts. Start here.

• Pre-wired PARA structure
• Notion capture dashboard
• AI synthesis prompt pack
• This guide — always free

Download the Vault →

The AI Builders Workshop — $47/month

Build real AI systems around your own thinking — with Claude + Obsidian, live, no code.

• Live workshops · 1–2 per month, 60 minutes, hands-on
• The replay library · every workshop, kept forever
• The FAQ thread · ask anything between sessions
• Templates from each session · copy, don’t rebuild
• The builders room · serious people, no hype, no gurus

$47/month founding price for the first 50 builders — then the price climbs until it settles at the standing rate of $97. You keep whatever price you join at. Cancel anytime. This window won’t be repeated.

See the Workshop →

The Bridge Install — By Application

Your AI chief of staff, vault, and publishing pipeline, built in YOUR business in 2–3 weeks, documented, yours. Two per month.

• Built in your business, not a template
• Your CLAUDE.md written to your voice
• The full system installed and documented
• The keys handed over — you own it

Apply for the Install →