How We Use Obsidian as Our Agency’s Second Brain (And Why Our AI Never Forgets a Client)

by
Apr 04, 2026
Posted in: AI & Web Development
Obsidian second brain for agency workflow with connected notes and AI

If you run an agency, you know the drill. Client info is scattered across Google Docs, old Slack messages, and whatever you half-remember from the last call. We were the same way until we figured out something better.

We use Obsidian as our central knowledge base. We connected it to Claude through an Obsidian MCP server, which means our AI can search, read, update, and create notes on its own. No copy-pasting. No re-explaining context every session. It just knows.

If you’re an agency owner or freelancer and you’re tired of re-teaching your AI who your clients are every time you open a new chat, this is the workflow we wish we had a year ago.

What Is Obsidian?

Obsidian is a free note-taking app. Everything is stored as plain Markdown files on your computer. No proprietary database. No cloud dependency. Just .md files in folders you own.

What makes it different from Notion or Google Docs:

  • Local-first – notes live on your machine, not someone else’s server
  • Markdown-based – every note is a simple text file you can open anywhere
  • Bidirectional linking – link notes together with [[double brackets]] and Obsidian maps the relationships for you
  • Tags – filter and organize notes across your whole vault with hashtags
  • Plugin ecosystem – hundreds of community plugins that extend what it can do
  • Graph view – visual map of how all your notes connect

As an Obsidian note-taking workflow, it’s great for personal use. For running an agency with AI built into daily operations, it’s on another level. But only once you connect it to something that can actually use those notes.

The Problem: AI Doesn’t Remember Anything

Here’s the thing about working with AI. Every conversation starts from zero. Claude doesn’t remember what you talked about yesterday. It doesn’t know who your client is, what their site looks like, what keywords you’re going after, or what the plan was.

So you’ve got three options:

  1. Re-explain everything every session. Waste 10-15 minutes setting context before you do any real work.
  2. Keep a .md file in your project folder and paste it in manually. Better, but you’re still maintaining those files by hand.
  3. Build a knowledge base the AI can access directly.

We went with option 3. It changed everything.

Comparison of AI workflow without Obsidian vs with Obsidian and MCP server

How We Connect Obsidian to Claude With an MCP Server

The connection runs through an MCP server (Model Context Protocol). MCP is an open standard that lets AI tools plug into external data sources. The Obsidian MCP setup uses two pieces working together to give Claude direct access to our vault.

On the Obsidian side, we use the Local REST API community plugin by Adam Coddington. It runs a local API inside Obsidian that exposes your vault, notes, tags, search, all of it, through API endpoints.

On the Claude side, we use obsidian-mcp-server, the Obsidian MCP server that connects to that API. Once connected, Claude can:

  • Search notes by name, tag, or content
  • Read any note in the vault
  • Create new notes from scratch
  • Update existing notes (append, prepend, or full overwrite)
  • Follow links between connected notes

This works through Claude Code and Claude Cowork. The Claude Code Obsidian integration means that when we start a session, Claude already has access to every note. We don’t paste anything in. We don’t upload files. We just say “check the client note” and it pulls it up.

Obsidian MCP server flow diagram showing vault connecting to Claude Code

How We Build a Client’s Main Note

When we onboard a new client, or when we want to document what we already know about an existing one, we have Claude scan their entire website and build what we call the main note.

The process is straightforward:

  1. We point Claude at the client’s website. Homepage, about page, services, team page, anything relevant.
  2. Claude pulls out everything. Who they are, what they do, where they’re located, their team, services, notable clients, what makes them different.
  3. Claude writes a structured Obsidian note with all of that organized into clean sections.
  4. We review it, adjust anything that needs adjusting, and it lives in the vault from that point forward.
How we build a client main note in Obsidian - 4 step process from website scan to structured vault note

That main note becomes the source of truth for that client. Every future session, whether it’s an SEO audit, a content strategy session, or a site redesign, Claude reads that note first. It already knows who the client is before we say a word.

No more spending the first 10 minutes of a session explaining the client’s business.

What Goes in a Main Note

A typical client main note includes:

  • Company overview – name, year founded, location, contact info
  • What they do – core services, specializations, industries
  • Team – key people, roles, bios
  • Service area – geographic focus, target markets
  • Notable clients or work – reference points and social proof
  • Differentiators – what sets them apart from competitors
  • Links to related notes – SEO plans, keyword lists, content calendars

Every client note gets a unique tag like #clientname so Claude can pull up everything related to that client across the entire vault instantly.

Session Notes: Where the Real Work Happens

The main note is the foundation. The real value shows up in the notes we create during work sessions.

For one of our Houston law firm clients, the vault might look like this:

  • Main note – company overview, services, team, everything about who they are
  • SEO and link building plan – full keyword analysis, page-by-page breakdown, blog topics, 12-month strategy with budget
  • Keyword research – target keywords per page, search volumes, difficulty scores
  • Content calendar – blog topics mapped to content pillars
  • Technical audit notes – site speed, Core Web Vitals, fixes needed
  • Meeting notes – decisions, action items, preferences

All of these are tagged and linked. The SEO plan links back to the main note. The keyword research links to the SEO plan. The content calendar references the keyword research. Everything is connected.

When we start a new session, Claude reads the main note to remember who the client is, then pulls up whatever task note is relevant. We pick up exactly where we left off. Nothing gets lost. Nothing gets repeated.

Obsidian vault structure showing main client note linked to SEO plan, keyword research, content calendar, and meeting notes

How We Organize Our Vault

We’re still figuring this out honestly. But here’s what works for us right now:

📁 Clients
   📁 Client A
   📁 Client B
   📁 Client C
📁 Muze Development
📁 SEO
📁 Things to look into
📁 Tips

Clients is the main hub. Each client gets their own subfolder with the main note and all task notes inside.

Muze Development is our own company notes. Same main note structure we build for clients, but for us. Our SEO plan, link building strategy, keyword research. We use the same system on ourselves.

SEO has general research and frameworks that aren’t tied to a specific client.

Tips and Things to look into are basically scratch pads for ideas and tools we want to explore later.

This will probably look different six months from now. That’s fine. Since everything in Obsidian is just Markdown files in folders, reorganizing is drag and drop. Links update automatically. Nothing breaks.

Tags and Links: How It All Stays Connected

Two Obsidian features make this whole system work.

Tags

Every note gets at least one tag. Client notes get a client-specific tag. When Claude needs everything related to a client, it searches by tag. Main note, SEO plan, keyword list, meeting notes, all of it comes up.

Bidirectional Links

Every task note links back to the client’s main note using [[wiki-link]] syntax. The main note links forward to all related task notes. This creates a web of connected info that Claude can navigate naturally.

When we say “check the SEO plan,” Claude follows the link from the main note, reads the plan, and picks up the conversation with full context. When we create a new note during a session, Claude links it back to the main note and tags it. The vault grows on its own without us maintaining some kind of master directory.

What We Actually Use This For

This isn’t hypothetical. Here’s what’s in our vault right now:

For Muze Development (us):

  • Company overview and positioning
  • Full SEO audit and link building strategy broken into monthly phases
  • Page-by-page keyword analysis for every page on our site
  • Blog topic ideas organized by pillar
  • Growth strategy and future plans

For our law firm clients:

  • Firm overview notes with practice areas, attorneys, service areas
  • Local SEO keyword research and tracking
  • Content strategies for personal injury, family law, and estate planning practices
  • Technical audit findings and fix tracking
  • Competitor analysis

General reference:

  • SEO frameworks and checklists
  • WordPress development best practices
  • Tool evaluations
  • Business development ideas

Every note is something Claude can access. Every session starts with context instead of a blank slate.

Why This Matters

If you’re using AI in your agency workflow, the quality of your output comes down to the quality of your context. That’s it. An Obsidian AI workflow solves this by giving your assistant persistent memory.

AI with no context writes generic stuff. AI that knows the client’s business, their keywords, their competitors, their goals, and the strategy you’ve been building for months writes stuff that actually moves the needle.

That’s the difference between AI as a fancy autocomplete and AI as an actual team member.

It Compounds

Every note makes the vault more valuable. Month one, you have a main note and a keyword list. Three months in, you have SEO plans, content calendars, audit results, and meeting notes all cross-referenced. Six months in, Claude can trace the entire history of a client engagement without you saying anything.

That’s the Obsidian second brain concept in practice. Not just for you, but for your AI. The vault remembers so you don’t have to. And more importantly, so your AI doesn’t have to guess.

What This Replaced For Us

Before this setup, here’s how it went:

  • Start a Claude session
  • Spend 10 minutes explaining the client
  • Paste old notes from a Google Doc
  • Realize we forgot half the context
  • Get mediocre output because the AI was working with incomplete info
  • Do it again next session

Now:

  • Start a Claude session
  • Say “read the main note for [client]”
  • Claude already knows everything
  • Do real work immediately

The time savings alone would justify the setup. But the quality difference in output is where it really pays off.

FAQ

Is Obsidian free?
The desktop and mobile apps are both free. We pay $4/month for Obsidian Sync to keep our vault synced between desktop and iPhone with end-to-end encryption, but that’s optional. You can sync through iCloud or Dropbox for free. Everything we described in this post works with the free version.

Do I need to know Markdown?
Not really. Obsidian has a visual editor mode. If you can write a heading and a bullet list, you’re good. Claude handles most of the formatting when it creates or updates notes anyway.

Is my data private?
Yes. Everything is stored locally on your machine. Nothing goes to Obsidian’s servers unless you opt into their Sync service. When Claude reads notes through MCP, it’s reading from your local files.

Can I use this with other AI tools?
MCP is an open protocol and other AI tools are starting to support it. Our Obsidian Claude Code workflow uses Claude Code and Claude Cowork specifically, but the concept applies to any AI tool that supports MCP servers.

How long does setup take?
Creating the vault takes a few minutes. You’ll need to install the Local REST API plugin inside Obsidian, then set up the Obsidian MCP server in Claude Code. Takes some terminal comfort, but it’s a one-time thing. Under an hour if you follow the docs.

What if I have a lot of clients?
It scales fine. Each client gets a subfolder and a tag. Claude searches by tag, so 3 clients or 30, it finds the right notes instantly.

Can my team use the same vault?
Yes. Share the vault folder through cloud storage (Dropbox, iCloud, Google Drive) or use Obsidian’s paid Sync feature.

Key Takeaways

  • Obsidian is free and local-first. Your notes are Markdown files on your machine that you own. Obsidian note taking is built on open formats you control.
  • The MCP server connection lets Claude read, create, update, and search your notes directly. No copy-pasting.
  • Build a main note for every client by having Claude scan their website and pull out key details into a structured reference doc.
  • Create task-specific notes for SEO plans, keyword research, content calendars, audits. Link them all back to the main note.
  • Tags and links keep everything connected so Claude can find anything in seconds.
  • Start simple. Your vault structure will change as you figure out what works.
  • It compounds. Every note you add makes future sessions faster and output better.
  • Context is everything. AI with full client context produces dramatically better work than AI starting from zero.

Resources

We’re Muze Development, a WordPress agency in Houston. We build custom websites and handle SEO for law firms and professional services companies across Texas. AI is built into how we develop, optimize, and manage client projects. If you want to see how we work, check out our services or reach out.

Subscribe to our blog for more on how we use AI, WordPress, and SEO to build better websites.

Newsletter Signup

Name(Required)
Email(Required)
Privacy(Required)