CyrilXBT on X: "Your Obsidian Vault Can Now Write Back to Itself. Here's the Architecture Nobody's Talking About" / X

Original URL: https://x.com/cyrilxbt/status/2052923836090167526?s=48

Title: CyrilXBT on X: "Your Obsidian Vault Can Now Write Back to Itself. Here's the Architecture Nobody's Talking About" / X

URL Source: http://x.com/cyrilxbt/status/2052923836090167526?s=48

Markdown Content: Most people use Obsidian as a one-way system.

Information goes in. Notes get created. Files get saved. The vault grows.

And that is where the intelligence stops.

You add to the vault. The vault never adds back.

That model is about to feel as outdated as manually checking email on a schedule.

Because with the right architecture connecting Claude Code to Obsidian via MCP, your vault is no longer a passive repository waiting for you to deposit something.

It is an active intelligence system that reads itself, processes what it finds, draws connections you never explicitly made, generates new documents from existing ones, updates its own records based on external events, and deposits the output back into the correct folder without you initiating any of it.

Your vault writes back to itself.

This article is the full architecture of how that works.

Before you understand the architecture you need to understand the shift in mental model it requires.

The default relationship with a note-taking app is archival.

Something happens. You write it down. You file it. You search for it later. Maybe.

The relationship with a self-writing vault is operational.

Something happens. The system reads about it. The system processes it against everything it already knows. The system produces an output. The output gets filed in the right place automatically. The next time something related happens the system has richer context to work with.

This is not a marginal improvement over how most people use Obsidian.

It is a categorical upgrade in what a personal knowledge system can do.

The people who build this architecture are not just taking better notes.

They are running a personal intelligence operation that compounds every week without additional effort.

Three layers make this architecture work.

Layer 1 is the knowledge layer. This is Obsidian itself. Your plain text Markdown files, organized in a structure that makes their content machine-readable and their relationships navigable.

Layer 2 is the connection layer. This is MCP (Model Context Protocol) — the bridge that gives Claude Code direct read and write access to every file in your vault in real time. Not through a clipboard. Not through copying and pasting. Direct file system access that lets Claude create, modify, and delete notes the same way you do.

Layer 3 is the intelligence layer. This is Claude Code operating through a carefully designed system of agent workflows that run on schedule or on trigger, read specific parts of the vault, process what they find, and write the results back.

Each layer is necessary. Remove any one of them and the system collapses into something much simpler.

Obsidian without MCP is just files.

MCP without Claude is just file access.

Claude without Obsidian structure is just a chat session.

Together they are something none of the three is individually.

The vault structure is the foundation that makes autonomous writing reliable.

If your vault is a disorganized collection of files with inconsistent naming and no clear taxonomy, Claude cannot reliably navigate it, and autonomous write-back operations deposit outputs in unpredictable places.

The structure that works best for a self-writing vault is a modified PARA system with two additional folders that the autonomous system uses as its primary operating surface.

00 - Inbox/ 01 - Projects/ [project-name]/ overview.md notes/ outputs/ 02 - Areas/ 03 - Resources/ 04 - Archive/ 05 - System/ CLAUDE.md Skills/ Templates/ 06 - Daily Notes/ 07 - Generated/ ← NEW: where autonomous outputs land 08 - Queue/ ← NEW: where pending tasks for Claude live

The Generated folder is where Claude deposits everything it produces autonomously. Weekly syntheses. Research briefs. Connection maps. Retrospective summaries. All of it lands here, dated and tagged, ready for your review.

The Queue folder is how you communicate tasks to Claude without initiating a session. You drop a file in Queue with a description of what you want. Claude reads the Queue on a schedule, processes the request, deposits the output in Generated, and moves the request file to Archive.

This asynchronous workflow is one of the most powerful aspects of the self-writing vault. You capture a task whenever you think of it. Claude executes it whenever its scheduled workflow runs. You review the output when it is convenient.

You are never waiting for Claude. Claude is never waiting for you.

The MCP connection is what transforms Claude from a chat interface into a vault operator.

The configuration file that makes this work:

{ "mcpServers": { "obsidian-vault": { "command": "npx", "args": [ "-y", "

", "/path/to/your/obsidian/vault" ] } } }

Replace /path/to/your/obsidian/vault with the actual path to your vault on your machine. On Mac this is typically /Users/[username]/Documents/[vault-name]. On Windows it is typically C:/Users/[username]/Documents/[vault-name].

Save this as claude_desktop_config.json in the appropriate directory. On Mac: ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows: %APPDATA%/Claude/claude_desktop_config.json.

Restart Claude Desktop.

Verify the connection with this prompt:

Read the CLAUDE.md file from my vault and list the top-level folders you can see. Tell me the total number of files in the vault and identify the most recently modified file.

If Claude returns accurate information about your vault, the connection is live and the intelligence layer can operate.

This is where the vault starts writing back to itself.

Six workflows power the self-writing vault architecture. Each one reads something, processes it, and writes back.

Workflow 1: The Daily Context Generator

Every morning at 6AM this workflow fires automatically via N8N.

It reads your most recent daily note, your active project files, and your inbox folder. It synthesizes what it finds into a structured daily context note and deposits it in Generated with the filename DAILY-CONTEXT-YYYY-MM-DD.md.

The daily context note contains:

A synthesis of where each active project stands based on the most recent notes in each project folder.

A list of open loops from the previous 7 days of daily notes that have not been resolved.

Connections between recent captures that point to the same underlying theme.

One suggested focus for the day based on deadlines, momentum, and open loops.

You open your laptop. The context note is already there. You do not compile it. You do not read through yesterday's notes. The vault synthesized them for you while you slept.

The Claude prompt that powers this:

Read the following files from my Obsidian vault: 1. The most recent daily note in 06 - Daily Notes/ 2. All overview.md files in 01 - Projects/ 3. All files added to 00 - Inbox/ in the last 48 hours Synthesize what you find into a daily context note with these sections: - PROJECT STATUS: Current state of each active project in 2-3 sentences - OPEN LOOPS: Unresolved items from the last 7 daily notes - EMERGING PATTERNS: Themes appearing across multiple recent notes - SUGGESTED FOCUS: One specific thing to work on today with reasoning Save the output as 07 - Generated/DAILY-CONTEXT-[TODAY'S DATE].md Format the filename as DAILY-CONTEXT-YYYY-MM-DD.md

Workflow 2: The Connection Finder

Once a week this workflow reads every note created in the past 7 days and finds non-obvious connections to older notes in the vault.

This is the workflow that makes the vault genuinely intelligent.

Most connection-finding in Obsidian is manual. You write a note and you manually decide what to link it to based on what you remember exists. The connections you do not make are often the most valuable ones.

The Connection Finder reads your new notes, searches the vault for related content, and produces a connection report that surfaces the links you did not make:

Read all files created or modified in the last 7 days in my Obsidian vault. For each file, search the entire vault for content that: 1. Addresses the same problem from a different angle 2. Contains a contradictory position worth noting 3. Represents the same principle applied in a different domain 4. Provides evidence for or against a claim made in the new file Do not surface obvious connections. Only surface non-obvious ones. For each connection found: - Source note: [filename and relevant excerpt] - Connected note: [filename and relevant excerpt] - Connection type: [same problem / contradiction / cross-domain / evidence] - Why this matters: [one sentence] Save the output as 07 - Generated/CONNECTIONS-[DATE].md

Workflow 3: The Queue Processor

This workflow runs every 2 hours and checks the 08 - Queue folder for any files you have dropped there.

If it finds files, it processes each one, deposits the output in 07 - Generated, and moves the processed request to 04 - Archive.

The Queue format is simple. Each file is named with a verb and a topic:

RESEARCH-quantum-computing-applications.md

SYNTHESIZE-all-notes-on-content-strategy.md

DRAFT-article-on-MCP-architecture.md

ANALYZE-my-project-timeline-this-quarter.md

Claude reads the filename to understand the task type and the file content for specific instructions.

The Queue system is the most flexible part of the self-writing vault because it lets you assign any task to the vault at any time without opening Claude or initiating a session.

You think of something at midnight. You drop a file in Queue. Claude processes it at 2AM. The output is waiting when you wake up.

Workflow 4: The Weekly Synthesis

Every Sunday at 8PM this workflow reads the entire week of activity and generates a structured retrospective.

It reads all daily notes from the past 7 days, all files modified this week across active projects, all outputs Claude generated this week in the Generated folder, and any files archived this week from the Queue.

The output is a weekly synthesis note that answers four questions:

What moved forward this week and what specifically caused it.

What did not move and what is the honest reason.

What pattern appeared across multiple days or contexts that points to something worth paying attention to.

What the three most important things to focus on next week are, ranked by leverage.

The weekly synthesis becomes its own valuable vault entry. Over time these notes become a documented record of how your thinking and work evolved. Claude can read them when doing longer-term analysis.

Workflow 5: The Project Auto-Updater

This workflow watches for changes in your project folders.

When a new note is added to a project subfolder, when a file is modified in a project, or when a new file appears in the project's inbox, this workflow fires.

It reads the change, reads the existing project overview, and updates the project overview with a brief note about what changed and what it means for the project status.

Your project overviews stay current automatically.

You never have to manually update a project status document. The vault updates it every time something new happens in that project.

Workflow 6: The Knowledge Distillation Engine

This workflow runs monthly and is the most computationally intensive of the six.

It reads every note in your Resources and Areas folders that has been modified or added in the past 30 days.

For each group of related notes it produces a distillation: the key insights compressed into a single well-structured reference document that synthesizes everything across the source notes without repeating content between them.

These distillations become the permanent, refined knowledge layer of your vault.

Raw notes accumulate detail and context.

Distillations accumulate insight and clarity.

The combination of both is more valuable than either alone.

Every autonomous workflow reads your CLAUDE.md before executing any task.

This file is the constitution of your self-writing vault. It tells Claude everything it needs to know to operate intelligently rather than generically.

CLAUDE.md — Vault Intelligence System ## Owner [Your name, role, primary focus areas] ## Vault Structure 00 - Inbox: Unprocessed captures 01 - Projects: Active projects with one folder per project 02 - Areas: Ongoing responsibilities 03 - Resources: Reference material 04 - Archive: Completed and outdated content 05 - System: Configuration files and templates 06 - Daily Notes: Daily journal entries 07 - Generated: Autonomous outputs — DO NOT manually edit 08 - Queue: Pending tasks for autonomous processing ## Active Projects [List each active project with a one-sentence status] ## My Voice [Description of how you write, what you value, your characteristic patterns] ## What Matters Most Right Now [Current priorities that should weight every analysis] ## Hard Rules for Autonomous Operations - Never delete any file without explicit instruction in the task - Never modify files in 05 - System without explicit instruction - Always append today's date to generated filenames - When uncertain about placement, deposit in 07 - Generated - Log every write operation in 07 - Generated/OPERATIONS-LOG.md ## Quality Standard for Generated Content [What good output looks like in your vault]

The Hard Rules section is the most important. Autonomous write operations without clear constraints produce unpredictable results. The rules make every operation predictable and auditable.

The six workflows need a trigger system to run automatically.

N8N is the best option for self-hosted automation. A $5 per month DigitalOcean droplet running N8N handles all six workflows with no per-execution pricing.

Each workflow in N8N has the same basic structure:

A Cron trigger (or a File Watch trigger for the Queue Processor and Project Auto-Updater).

An HTTP Request node that calls the Claude API with the appropriate system prompt.

A File Write node that saves the output to the correct location in your vault.

An optional Notification node that sends you a message when the workflow completes.

The N8N workflow template for a scheduled vault operation:

Cron Trigger → HTTP Request (Claude API) → Write File → Slack/Telegram Notification

The Claude API call includes the full system prompt for that workflow plus the CLAUDE.md contents as context.

The compounding effect of a self-writing vault is not immediately obvious.

In week one it feels like a productivity tool. Slightly more context in the morning. A few interesting connections surfaced. Some queue tasks processed overnight.

By week four it starts feeling different.

The daily context notes contain references to patterns Claude noticed three weeks ago that are now reappearing. The connection reports are surfacing links between notes you had genuinely forgotten existed. The project overviews are more current and accurate than they ever were when you were maintaining them manually.

By month three the vault is operating with context that took months to accumulate.

Every autonomous workflow benefits from every note that has ever been written.

Every distillation incorporates every insight from every piece of related material.

Every connection report gets richer because there is more to connect.

The vault does not just write back to itself.

It learns.

Not in the machine learning sense.

In the practical sense that the outputs it generates in month three are meaningfully more useful than the outputs it generated in week one because it has three months of your thinking to work with.

The self-writing vault architecture is not primarily a productivity tool.

It is a thinking amplifier.

The reason most people do not make the connections their notes contain is not laziness or lack of intelligence.

It is cognitive bandwidth.

You cannot hold 2,000 notes in working memory simultaneously. You cannot remember that the insight you wrote 8 months ago is directly relevant to the problem you are working on today.

Claude can.

Claude reads all 2,000 notes every time it runs the Connection Finder workflow. It holds all of them in context simultaneously. It finds the link between the note you wrote 8 months ago and the problem you are working on today without needing to remember that either existed.

The self-writing vault does not replace your thinking.

It extends the range of what your thinking can connect.

And that extension compounds every week you use it.

Build the architecture this weekend.

Start with the Daily Context Generator and the Queue Processor.

Add the Connection Finder in week two.

Add the rest in week three.

By month two you will have a vault that is more useful than any static note-taking system you have ever used.

By month six you will not remember how you thought without it.

Follow

for the exact N8N workflows, Claude prompts, and MCP configurations that make this entire architecture run.