Claude Code + Obsidian: Building an AI-Powered Second Brain

A second brain captures your knowledge. An AI-powered second brain helps you use it.

The combination of Claude Code and Obsidian creates something new: a knowledge system that you can converse with, query, and synthesize from. Instead of searching through notes trying to remember where you wrote something, you ask Claude to find it, summarize it, and connect it to your current thinking.

This guide covers how to set up Claude Code with Obsidian and the workflows that make this combination powerful.

The Vision: AI + PKM

Traditional PKM (Personal Knowledge Management) has a retrieval problem. You capture information diligently, but finding the right note at the right time requires:

  • Remembering that you captured it
  • Remembering where you put it
  • Constructing the right search query
  • Reading through results to find the relevant section

AI changes this equation. Instead of searching, you describe what you need:

“What did I learn about pricing strategies from my customer interviews last quarter?”

Claude reads your notes, synthesizes the relevant information, and presents a coherent answer with citations.

This isn’t futuristic speculation. It works today with Claude Code and Obsidian.

How It Works

Claude Code has direct access to your file system. Point it at your Obsidian vault, and it can:

  • Read any note
  • Search across files
  • Understand Markdown formatting
  • Parse frontmatter metadata
  • Follow links between notes

Unlike web-based AI tools, Claude Code runs locally with full access to your files. No uploading, no size limits (beyond context windows), no privacy concerns.

Setting Up Claude Code with Obsidian

Step 1: Locate Your Vault

Find your Obsidian vault’s file path. On macOS, it’s typically:

~/Documents/Obsidian/YourVaultName

or wherever you chose to create it.

Step 2: Create a CLAUDE.md File

Add a CLAUDE.md file to your vault root. This gives Claude context about your system:

# Obsidian Vault: [Your Name]'s Second Brain

## Structure
- `/Projects` - Active work with deadlines
- `/Areas` - Ongoing responsibilities
- `/Resources` - Reference materials
- `/Archives` - Completed/inactive items
- `/Daily` - Daily notes (YYYY-MM-DD format)

## Conventions
- Notes use YAML frontmatter with: date, tags, type, status
- Links use [[wikilink]] syntax
- Tasks use `- [ ]` checkbox format
- Tags start with # and use kebab-case

## Key Tags
- #customer-feedback - Direct customer input
- #product-idea - Feature and product ideas
- #meeting - Meeting notes
- #research - Research and learning
- #decision - Important decisions made

## Important Notes
- [[Product Roadmap]] - Current product direction
- [[Customer Personas]] - Target user profiles
- [[Competitive Analysis]] - Market landscape

Step 3: Open Your Vault in Claude Code

Navigate to your vault directory:

cd ~/Documents/Obsidian/YourVaultName
claude

Claude Code now has access to all your notes.

Powerful Workflows

Research Synthesis

You’ve accumulated notes from articles, books, podcasts, and conversations. Claude can synthesize them:

Prompt:

“Read all my notes tagged #pricing-strategy and summarize the key frameworks and insights. Cite which notes each insight comes from.”

Claude’s response will pull from multiple notes, identify patterns, and present a coherent summary with references to specific files.

Finding Connections

Your vault contains hundreds of notes. Connections exist that you haven’t noticed:

Prompt:

“What connections exist between my notes on customer acquisition and my notes on product development? Are there insights from one domain that apply to the other?”

Claude reads both sets of notes and identifies overlapping themes, contradictions, or opportunities.

Content Creation from Notes

Turn research into output:

Prompt:

“Using my notes in /Resources/SEO-Knowledge, draft a 1500-word blog post outline about technical SEO for startups. Reference specific notes where relevant.”

Claude transforms your accumulated knowledge into structured content, citing your own notes as sources.

Decision Support

When facing a decision, query your past thinking:

Prompt:

“I’m deciding whether to raise funding or stay bootstrapped. What have I captured in my notes about this topic? What do my previous thoughts and research suggest?”

Claude surfaces relevant notes from customer interviews, articles you’ve saved, and your own reflections.

Weekly Review Automation

Streamline your weekly review:

Prompt:

“Summarize my daily notes from this week. List all tasks marked complete, all tasks still open, and any recurring themes or concerns.”

Claude parses your daily notes and generates a review summary.

Structuring Notes for AI

Your vault works better with Claude when notes follow consistent patterns.

Use Frontmatter

YAML frontmatter gives Claude structured data to work with:

---
date: 2025-01-15
type: meeting
tags: [customer-feedback, product]
attendees: [John Smith, Sarah Chen]
company: Acme Corp
status: processed
---

Claude can filter and sort by these fields:

“Show me all meeting notes with Acme Corp from the last 60 days.”

Consistent Tagging

Tags create cross-cutting categories that Claude can query:

  • #insight - Key realizations worth remembering
  • #question - Open questions to investigate
  • #decision - Choices made and reasoning
  • #follow-up - Items needing action

Prompt:

“What open questions (tagged #question) do I have across my vault? Group them by topic.”

Atomic Notes Help AI

Small, focused notes are easier for Claude to understand and cite than long, multi-topic documents. When Claude references “the third paragraph of your 50-page meeting notes,” that’s less useful than referencing a specific note titled “Customer Feedback: Pricing Concerns - Jan 2025.”

Links help Claude understand relationships:

“What notes link to [[Product Roadmap]]? What are the main themes among those linking notes?”

Real Examples

Example 1: Investor Pitch Preparation

Context: You’re preparing for investor meetings and want to gather relevant information from your vault.

Prompt:

“I’m preparing for investor meetings next week. Read my notes and compile:

  1. Key metrics and traction data from my /Projects/Metrics notes
  2. Customer testimonials and feedback from #customer-feedback notes
  3. Market size data from /Resources/Market-Research
  4. Our competitive advantages from [[Competitive Analysis]] Format as a briefing document I can review before meetings.”

Output: A comprehensive briefing pulling from multiple vault locations.

Example 2: Hiring Decision

Context: You’re deciding whether to hire a specific role.

Prompt:

“Search my notes for anything related to hiring, team building, or the [specific role] function. What have I learned about:

  • When to hire for this role
  • What to look for in candidates
  • Mistakes to avoid Include any relevant insights from books or articles I’ve captured.”

Output: Aggregated wisdom from your notes on the topic.

Example 3: Product Prioritization

Context: You need to decide what to build next.

Prompt:

“Analyze my #customer-feedback notes from the last 90 days. What are the top 5 requested features or most common pain points? For each, estimate how many different customers mentioned it and summarize their specific concerns.”

Output: A data-driven prioritization input based on your own customer research.

Limitations and Best Practices

Context Window Limits

Claude has a context window (how much text it can process at once). Large vaults may exceed this.

Solutions:

  • Be specific about which folders or tags to search
  • Use date ranges to limit scope
  • Process in batches for vault-wide analysis
  • Structure requests to focus on relevant subsets

Stale Information

Your notes may contain outdated information. Claude doesn’t know what’s current.

Solution: Use frontmatter dates and ask Claude to prioritize recent notes:

“Focus on notes from 2025. If you find conflicting information, prefer the more recent note.”

Privacy Considerations

Claude Code processes files locally, but if you’re using API-based Claude, your notes are sent to Anthropic’s servers.

Solutions:

  • Use Claude Code’s local mode where possible
  • Avoid querying notes with highly sensitive information
  • Consider what you capture in your vault

Over-Reliance Risk

AI retrieval is powerful but not perfect. Don’t abandon manual note review entirely.

Balance: Use AI for synthesis and search. Continue regular manual review for serendipitous discovery and deeper understanding.

The Future: MCP and Deeper Integration

Model Context Protocol (MCP) enables standardized connections between AI models and external tools. Obsidian MCP servers are emerging that provide:

  • Real-time vault updates as context
  • Structured access to Obsidian’s graph
  • Integration with Obsidian plugins
  • Bidirectional communication (AI can suggest links, create notes)

This is early-stage but points toward tighter integration between AI assistants and knowledge management tools.

Getting Started

This Week

  1. Install Claude Code if you haven’t
  2. Create a CLAUDE.md file in your vault
  3. Open your vault in Claude Code
  4. Try one query: “Summarize the main themes in my recent notes”

This Month

  1. Add consistent frontmatter to new notes
  2. Develop a tagging taxonomy
  3. Experiment with synthesis prompts
  4. Create a weekly review prompt you reuse

Ongoing

  1. Refine your CLAUDE.md as your system evolves
  2. Build a library of useful prompts
  3. Balance AI retrieval with manual exploration
  4. Stay current with MCP developments

Bottom Line

Obsidian captures your knowledge. Claude helps you use it.

The combination transforms a static note collection into a dynamic knowledge partner. Ask questions, get synthesized answers, and surface connections you’d never find manually.

This isn’t about replacing your thinking—it’s about augmenting it. Your notes become more valuable when you can actually access and apply what they contain.

Start small. One query at a time. Let the value reveal itself through use.

Resources