I Tried Obsidian + Claude Code — Personally, I Couldn't Justify Installing It
Lately I’ve been seeing “Obsidian + Claude Code is a strong combo” everywhere.
I was curious, so I tried it myself.
Going in, I was skeptical: Obsidian is a note tool for humans — what’s actually supposed to change when you pair it with AI coding?
After running it, I personally couldn’t justify keeping it installed. Here’s how I got there.
The Setup: There’s a Real Persistence Problem
If you keep doing AI-assisted coding, you eventually hit the same wall:
“Where do we put long-term knowledge?”
The first instinct is to dump everything into CLAUDE.md, Skills, and Rules. That breaks down fast.
There’s just too much information to load every time.
The rules you want applied on every run (coding conventions, deploy steps) and the information you only consult occasionally (meeting notes, investigation logs, past failures) shouldn’t live in the same place.
The real question is the next one:
“Does that ‘separate place’ actually have to be Obsidian?”
From Claude Code’s POV, Obsidian Is “Just a Folder of MD”
You see it immediately once you start using it. What Claude Code can read is the plain Markdown files under the vault. That’s it.
- Wikilinks
[[note-name]]→ just text. Claude Code doesn’t auto-traverse them - Canvas (
.canvas) → JSON, but the spatial structure doesn’t translate well to text - Plugin output (Dataview, Tasks, etc.) → invisible. Only the source MD is visible
- Graph view → obviously not visible
Caveat: I haven’t tested this, but apparently with some settings or plugins you can expose more of the vault to AI (e.g., switching Wikilinks to standard Markdown links, or using a plugin that materializes Dataview results to MD). This article is about the default setup.
From Claude Code’s perspective, an Obsidian vault is “just a directory of Markdown files,” nothing more.
Drop your .md files in docs/ or notes/ and Claude Code does the same job.
Obsidian Only Helps the Human Side
Wikilinks, graph view, quick switcher, daily notes — these lift the writer’s motivation and search experience, not AI reasoning.
If “Obsidian makes AI smarter” becomes the casual reading, people will pick the wrong tool for the wrong reason.
I Thought “AI Writes → Human Reads” Would Be the Saving Grace
Up to this point I’d been saying “Obsidian is for humans, not AI.”
But for the case where AI writes a lot of notes that humans read later, I expected Obsidian’s UI (graph, wikilinks, quick switcher) to actually shine.
For example:
- Have Claude Code clean up meeting notes / chat logs and write them to the vault
- Have Claude Code write investigation logs: “Tried X → got stuck on Y → solved with Z”
- Have Claude Code document a section of the codebase into the vault
The roles flip:
- AI = producer (writing)
- Human = consumer (reading)
Going into the experiment, that was where I was hopeful.
Then I Realized: AI Itself Replaces the UI
Running it, another thing became clear:
With Claude Code in the loop, AI itself substitutes for most of Obsidian’s human-side UI.
For example, in this very project: I asked “which file is the Obsidian article?” — Claude Code grepped and returned it. No quick switcher needed.
| Obsidian feature | Replaceable by asking AI |
|---|---|
| Quick switcher (fuzzy file finder) | ✓ “Where’s that note?” works |
| Full-text search | ✓ AI runs grep |
| Backlinks (reverse links) | ✓ “Which notes mention this?” |
| Metadata queries (e.g., “all draft notes”) | ✓ AI parses frontmatter |
| Graph view | ✗ Hard to replace visually |
So even in the AI writes → human reads direction — if the human is reading through AI, Obsidian’s UI isn’t really needed.
What’s left: visual overview via graph, and offline/mobile reading without AI access.
Personal Conclusion: I Couldn’t Justify Installing It
For me, honestly:
- AI reading direction → plain Markdown is enough
- Human reading direction → I just ask the AI; it finds it
- Graph view → cool to look at once, then I never use it
- Offline / mobile reading → not in my workflow
So I couldn’t justify keeping Obsidian installed for my own use case.
When people say “Obsidian + Claude Code is strong,” I suspect it’s one of two things:
- People who were already deep in Obsidian saying “and AI made it better”
- People not heavily using AI saying “Obsidian’s human UI is amazing”
For someone who uses Claude Code daily, the workflow tends to complete itself without needing Obsidian.
When It Might Still Be Worth It
- You already have an Obsidian habit (I don’t)
- You actually use the visual graph for navigation
- You read notes offline / on mobile, away from AI access
Outside of those, notes/ directory + Claude Code is probably enough.
Takeaways
- AI-assisted knowledge management really needs Markdown plus a sane directory structure
- “Obsidian + Claude Code = strong” — the AI side gets zero direct lift; Markdown is doing the work
- I expected “AI writes → human reads” to be the saving grace, but AI itself replaces the UI
- Conclusion: for Claude Code users,
notes/directory is usually enough - Worth installing only if you have visual / offline / habitual reasons
Obsidian is a good tool. But in an environment with Claude Code, most of its role gets quietly absorbed by the AI itself.
For me, at least, I couldn’t see a reason to keep it.