Using an Obsidian vault with Claude Code
Plain Markdown on disk is the reason this works at all. Everything else follows from that.
A vault is just Markdown files in a folder, which is exactly the format a coding agent reads best. Point Claude Code at the directory and your notes become context it can search, summarise and cross-reference while you work. The setup pays off when notes are short, plainly titled and full of the words you would actually search for. It stops paying off for anything you capture away from the machine, which is where the vault has always been weakest.
This page describes a workflow you run yourself with local files. Mindle is cloud-hosted rather than a local vault and has no plugin API, so it is a capture and retrieval layer beside that setup, not a replacement for it. Claude Code can reach Mindle notes through its MCP connector, but they are still not files on your disk.
Why a vault works as agent context
There is no clever integration behind this. An Obsidian vault is a directory of .md files, and a coding agent already reads directories of text files well. That is the entire mechanism.
It means your decisions, architecture notes and half-finished reasoning become something you can point at during a task instead of retyping from memory. The agent reads the note about why you rejected the queue-based approach, and you do not have to explain it again.
- Markdown is plain text, so nothing has to be exported or converted
- Filenames and headings act as the index
- Wiki links give the agent a path between related notes
- The vault stays usable on its own if you stop doing this tomorrow
Structuring notes so an agent can use them
The habits that make a vault good for an agent are mostly the habits that make it good for you, applied more strictly.
Short notes beat long ones, because a 3,000 word note about six topics matches every query and answers none. Descriptive titles beat clever ones. Writing the words you would search for, rather than the words that sound best, matters more here than in prose you only read yourself.
The other rule is to record the reasoning, not just the outcome. A note saying we chose Postgres is worth little. A note saying we chose Postgres because the write pattern was append-heavy and we needed range queries answers the question that comes up six months later.
- One idea per note, titled as plainly as possible
- Decisions with their reasons and the alternative you rejected
- Dates on anything that will age
- Link out to the related note rather than repeating its content
Where the workflow breaks down
Everything above assumes you are at the machine with the vault on it. Most thinking does not happen there.
The idea arrives on a walk, in a meeting, while reading on a phone. To reach the vault it has to survive a manual round trip: capture it somewhere else, remember later, open the vault, write it up, file it. That trip is where most notes die, and it is the failure mode a vault has had since long before agents existed.
The second gap is retrieval by meaning. Filename and full-text search find notes that share your words. They do not find the note from four months ago that made the same point in different vocabulary, which is usually the one worth having.
Where Mindle fits, and where it does not
Being direct about this is more useful than pretending otherwise. Mindle is cloud-hosted rather than a local vault of files you own, there is no plugin API, and there is no export path yet. Claude Code can read and write Mindle notes through the MCP connector, so the agent case is covered, but that is access over the network to a hosted account, not files on your disk. If your requirement is the files, keep the vault. That requirement is legitimate.
What Mindle covers is the part the vault has never covered: capture away from the machine, and retrieval that does not depend on matching your own past wording. Type it, say it, save the PDF, photograph the whiteboard, and it groups into concepts automatically by meaning rather than by the words it happens to contain.
Used together, the split is simple. The vault holds what you write at the desk, as files you own. Mindle holds everything you captured while away from it, and gives it back by subject. The agent can read both, each in its own way.
- Vault: local files, agent-readable, yours on disk
- Mindle: capture in any format, grouped and searchable by meaning
- No integration between them today, and this page is not claiming one
Questions
Does Claude Code need a plugin to read an Obsidian vault?
No. A vault is a folder of Markdown files, so pointing the agent at the directory is enough. That is why this workflow needs no integration on either side.
Does Mindle integrate with Claude Code?
Through MCP, yes. Add the connector, approve it once, and Claude Code can search, read, create and edit your Mindle notes. What does not exist is a plugin API or any local file access, so the vault half of this workflow stays as it is.
Can I keep my notes as local files in Mindle?
No. Mindle is cloud-hosted, with no local vault or file-per-note store, and there is no export path yet. If owning the files matters to you, an Obsidian vault is the better fit for that requirement.
What can I capture into Mindle?
Text and voice notes, files such as PDF, DOC, TXT and MD, and images. That range is the point: it covers the captures that never make it into a vault because you were not at the machine.
How should I write vault notes so an agent uses them well?
Keep them short and single-topic, title them plainly, and write down the reasoning behind a decision rather than only the decision. Broad notes covering many topics match everything and answer nothing.
Can I use both without duplicating everything?
Yes, if you split by where the thought happened. The vault holds what you write at the desk and what the agent reads. Mindle holds what you captured elsewhere, in whatever format was fastest at the time.
Keep reading
Keep the vault. Fix the capture gap.
Text, voice, files and images · Grouped by meaning · Mindle never trains AI on your notes
Try for free