Search years of research files in an Obsidian vault
2025-08-09 · ai · obsidian · productivity. Task: 2025-08-09 · ai · obsidian · productivity By Mauricio Gomes · 5 min read The Starting Point If you haven’t heard of Claude Code, it’s Anthropic’s command-line interface that gives you Claude with access to your filesystem. It can read and write files, run commands, and basically act as a coding assistant that can actually touch your computer. Obsidian, on the other hand, is my note-taking software. It’s where I store everything - daily notes, research papers, code snippets, random thoughts, scanned PDFs of old receipts, you name it. My vault has grown into this massive interconnected web of markdown files and documents that represents years of accumulated knowledge. Here’s the problem: every time I start a new Claude Code session (or Claude or ChatGPT), it’s a blank slate. It doesn’t know how I organize my notes. It doesn’t know my coding preferences. It doesn’t know that I have a massive collection of PDFs sitting in my Obsidian vault. It’s like having a brilliant assistant who gets amnesia every morning. The Solution: CLAUDE.md The magic happens with a single file called CLAUDE.md that lives in my Obsidian vault. Claude Code automatically reads this file at the start of every session, giving it context about who I am and how I work. This file tells Claude: How my Obsidian vault is organized My coding style preferences (like using Go’s any instead of interface{}) Where to find specific types of information How to interact with my note-taking system Custom commands and workflows I’ve developed The Knowledge Layer In my Obsidian vault, I’ve created a special folder with files that define who I am: A list of my interests (from Formula 1 to cellular automata) My movie collection (including physical media 😅) Research areas I’m actively exploring Even my preferred tools and workflows This means when I ask Claude to research something, it doesn’t just search generically - it searches with context. It knows I care about Raft consensus more than blockchain consensus. It knows I’m interested in arbitrage stories like when Michael Larson beat Press Your Luck in 1984. Searching My Digital Brain One of the coolest parts of this setup is teaching Claude to search through everything in my Obsidian vault - including PDFs. I’ve configured it to use macOS’s Spotlight (mdfind) to search not just my markdown notes but also all those research papers, scanned documents, and random PDFs I’ve collected over the years. Claude can now search through years of accumulated PDFs - research papers, scanned receipts, documentation, random articles I’ve saved. And with textutil, it can extract text from practically any document format: No more “sorry, I can’t read that format.” Everything is searchable, everything is accessible. Real Examples Daily Notes Without the Hassle Instead of navigating folder structures, I just say “add this to my notes” and Claude knows exactly where today’s note lives (~/work log/2025/2025-08/2025-08-09.md). It creates the folder structure if needed, follows my naming conventions, and just works. Personalized Research When I ask Claude to research a topic, it doesn’t just search the web. It knows my interests file, so when I ask about consensus algorithms, it focuses on Raft and Paxos (which I care about) rather than blockchain consensus (which I don’t). It’s like having a research assistant who’s read all my papers and knows my field. It even stores the results in my “research” folder. Finding That One PDF Remember that paper you scanned three years ago? The one about… something with quantum? Claude can find it: TV/Movie Recommendations If it knows everything I’ve watched and liked, it should be able to recommend new things. The Technical Setup Here’s what makes this work: CLAUDE.md: Lives at the root of my Obsidian vault, automatically loaded by Claude Code Context files: Stored in a special folder with my interests, preferences, and other personal data Spotlight integration: All documents in the vault are indexed and searchable Custom workflows: Defined in CLAUDE.md for common tasks The beauty is its simplicity. No APIs, no complex integrations. Just markdown files and bash commands. Why This Matters The real win isn’t just the productivity boost (though that’s nice). It’s the elimination of cognitive overhead. I don’t have to context-switch to “talking to a new AI” mode anymore. Claude knows: My conventions and preferences Where everything lives What I care about How I like to work; tools: Claude Code. Human role: The original post does not say how a person stayed involved.; limitations: This workflow is heavily dependent on macOS-specific tools (Spotlight/mdfind, textutil) and the local filesystem access of Claude Code. It requires a pre-existing, well-organized Obsidian vault and manual creation of context files. The source does not detail error handling for missing files or unsupported formats, nor does it specify how human approval is integrated into the automated note-taking or search processes..
Tools
Published on Agent Work Atlas:
Overview
2025-08-09 · ai · obsidian · productivity
Workflow
2025-08-09 · ai · obsidian · productivity By Mauricio Gomes · 5 min read The Starting Point If you haven’t heard of Claude Code, it’s Anthropic’s command-line interface that gives you Claude with access to your filesystem. It can read and write files, run commands, and basically act as a coding assistant that can actually touch your computer. Obsidian, on the other hand, is my note-taking software. It’s where I store everything - daily notes, research papers, code snippets, random thoughts, scanned PDFs of old receipts, you name it. My vault has grown into this massive interconnected web of markdown files and documents that represents years of accumulated knowledge. Here’s the problem: every time I start a new Claude Code session (or Claude or ChatGPT), it’s a blank slate. It doesn’t know how I organize my notes. It doesn’t know my coding preferences. It doesn’t know that I have a massive collection of PDFs sitting in my Obsidian vault. It’s like having a brilliant assistant who gets amnesia every morning. The Solution: CLAUDE.md The magic happens with a single file called CLAUDE.md that lives in my Obsidian vault. Claude Code automatically reads this file at the start of every session, giving it context about who I am and how I work. This file tells Claude: How my Obsidian vault is organized My coding style preferences (like using Go’s any instead of interface{}) Where to find specific types of information How to interact with my note-taking system Custom commands and workflows I’ve developed The Knowledge Layer In my Obsidian vault, I’ve created a special folder with files that define who I am: A list of my interests (from Formula 1 to cellular automata) My movie collection (including physical media 😅) Research areas I’m actively exploring Even my preferred tools and workflows This means when I ask Claude to research something, it doesn’t just search generically - it searches with context. It knows I care about Raft consensus more than blockchain consensus. It knows I’m interested in arbitrage stories like when Michael Larson beat Press Your Luck in 1984. Searching My Digital Brain One of the coolest parts of this setup is teaching Claude to search through everything in my Obsidian vault - including PDFs. I’ve configured it to use macOS’s Spotlight (mdfind) to search not just my markdown notes but also all those research papers, scanned documents, and random PDFs I’ve collected over the years. Claude can now search through years of accumulated PDFs - research papers, scanned receipts, documentation, random articles I’ve saved. And with textutil, it can extract text from practically any document format: No more “sorry, I can’t read that format.” Everything is searchable, everything is accessible. Real Examples Daily Notes Without the Hassle Instead of navigating folder structures, I just say “add this to my notes” and Claude knows exactly where today’s note lives (~/work log/2025/2025-08/2025-08-09.md). It creates the folder structure if needed, follows my naming conventions, and just works. Personalized Research When I ask Claude to research a topic, it doesn’t just search the web. It knows my interests file, so when I ask about consensus algorithms, it focuses on Raft and Paxos (which I care about) rather than blockchain consensus (which I don’t). It’s like having a research assistant who’s read all my papers and knows my field. It even stores the results in my “research” folder. Finding That One PDF Remember that paper you scanned three years ago? The one about… something with quantum? Claude can find it: TV/Movie Recommendations If it knows everything I’ve watched and liked, it should be able to recommend new things. The Technical Setup Here’s what makes this work: CLAUDE.md: Lives at the root of my Obsidian vault, automatically loaded by Claude Code Context files: Stored in a special folder with my interests, preferences, and other personal data Spotlight integration: All documents in the vault are indexed and searchable Custom workflows: Defined in CLAUDE.md for common tasks The beauty is its simplicity. No APIs, no complex integrations. Just markdown files and bash commands. Why This Matters The real win isn’t just the productivity boost (though that’s nice). It’s the elimination of cognitive overhead. I don’t have to context-switch to “talking to a new AI” mode anymore. Claude knows: My conventions and preferences Where everything lives What I care about How I like to work
Reported result
The source describes qualitative benefits such as reduced cognitive overhead and personalized research capabilities but does not provide measured metrics, success rates, or quantitative outcomes. It is unclear if the workflow consistently handles edge cases or errors without human intervention. A measurable acceptance check would be to verify that 100% of test queries for specific PDFs return the correct file path and extracted text snippet within a defined time limit.
Human role
The original post does not say how a person stayed involved.
Limitations
This workflow is heavily dependent on macOS-specific tools (Spotlight/mdfind, textutil) and the local filesystem access of Claude Code. It requires a pre-existing, well-organized Obsidian vault and manual creation of context files. The source does not detail error handling for missing files or unsupported formats, nor does it specify how human approval is integrated into the automated note-taking or search processes.
Try it
Act as a knowledge management assistant. I have an Obsidian vault at [replaceable vault path] and use Claude Code. First, create a CLAUDE.md file in the root that defines my folder structure for [replaceable note categories] and my interest in [replaceable topics]. Next, configure a search workflow using mdfind to locate PDFs in [replaceable PDF folder] and textutil to extract text. Perform a dry run: search for a document related to [replaceable keyword] and propose adding a summary to [replaceable daily note path]. Show me the proposed file changes and search results for my approval before executing any writes. If the search fails, report the error and ask for clarification on the file location. Do not modify any files without my explicit confirmation.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.