Claude Code: The Complete Guide for Knowledge Workers & Creators

Not a chatbot. Not autocomplete. An AI collaborator that reads your documents, edits your notes, runs your automations, and remembers your preferences — across every session, in any folder on your computer.

Claude Code is built for developers, but it runs the same way in a folder of Markdown notes as it does in a codebase. This guide covers what Claude Code actually is, how to set it up next to Obsidian, and the non-coding tasks — personal knowledge management, writing, research — it’s genuinely good at.

The Foundation

What Claude Code Actually Is

Most descriptions of Claude Code say something like “an AI coding assistant for developers.” That framing misses 80% of what it can do — and most of who it’s for. Here’s a more accurate description.

Claude Code is an AI collaborator that lives in your terminal. It can read every file in any folder you point it at. It can edit documents, create new files, reorganize your vault, summarize research, draft content in your voice, and remember your preferences across every session. Unlike Claude.ai (the browser chatbot), Claude Code doesn’t wait for you to paste content in. It goes and reads your actual files.

The “code” in the name is partly misleading. Yes, it was built partly for software development. But its underlying capability is much broader: it’s an AI that can act on your file system, connect to your tools, and execute multi-step tasks without you managing each step manually. That’s useful for anyone who works with a computer — not just programmers.

"Claude Code is the first AI tool I use that feels like a collaborator rather than a calculator."

The Key Insight: Claude Code Can Act, Not Just Respond

Every AI chatbot you’ve used before is reactive. You paste content in, it responds, you copy the response, you implement it yourself. The AI is an advisor. You are the executor.

Claude Code runs in agentic mode. You describe the outcome you want — in plain English — and Claude Code figures out what needs to happen, takes the actions, checks its own work, and tells you what it did. You describe the destination; Claude Code figures out the route.

Understanding the Difference

Claude Code vs. Claude.ai

Same underlying AI. Very different tools. Here’s the distinction that matters:

Claude.ai — The Chatbot

• Lives in your browser
• You paste content in manually
• Suggests text — you implement it yourself
• Starts fresh each conversation
• No access to your file system
• Great for quick Q&A and drafting

Claude Code — The Collaborator

• Lives in your terminal or desktop app
• Reads your actual files directly
• Edits, creates, and reorganizes files itself
• Remembers your preferences via CLAUDE.md
• Access to your full file system (with your permission)
• Handles multi-step autonomous tasks

The simplest way to think about it: Claude.ai is a conversation. Claude Code is a collaborator with access to your computer. Both are powerful. They’re not competing — they’re for different moments in your work.

The Surprising Answer

Who Claude Code Is Actually For

The name says “code.” The documentation talks about developers. But the people getting the most transformative use out of Claude Code are often not writing a single line of code. They’re using it to work differently with their writing, their notes, and their creative process.

Real people in the Claude Code community right now:

1. 01
   Knowledge Workers & PKM Practitioners Using Claude Code to query their Obsidian vaults in plain English, generate weekly reviews from Daily Notes, find orphaned notes, and build automation workflows that route captured ideas to the right place automatically.
2. 02
   Writers & Newsletter Creators Using Claude Code to analyze writing style from their own archives, draft newsletters from scattered notes, run content audits across their blog, and repurpose one piece of content into five different formats — all without copying and pasting between windows.
3. 03
   Researchers & Academics Querying hundreds of research notes in plain English ("what did I write about attention economics?"), cross-referencing sources, identifying gaps in literature, and synthesizing themes across large note archives.
4. 04
   Content Creators & Solopreneurs Building personal AI systems that know their voice, their products, and their audience — then delegating repetitive creative work to Claude Code while keeping creative direction for themselves.

The Use Case Most Guides Skip

Claude Code for Personal Knowledge Management

If you keep notes in Obsidian, Claude Code is the closest thing to a research assistant who has actually read your vault. Point it at your notes folder and it can query them in plain English, draft from what’s already there, and keep a running memory of your projects in CLAUDE.md — the same file it uses for any other kind of work.

This isn’t a separate product or a plugin you install. It’s the same terminal tool described above, aimed at a folder of Markdown notes instead of a codebase. Setup is below. For a full framework on running your vault this way — not just individual prompts — see the agentic PKM guide and how to build a second brain with Claude Code.

Costs & Plans

Pricing — What You Actually Need

Claude Code is not available on the free Claude.ai plan. Here are the tiers that include it:

Pro $20 / month

• Claude Code access
• Claude Sonnet 4.6 (default)
• ~40–80 hrs/week usage
• 5× free tier limits
• Best for: individuals starting out

Max 5× $100 / month

• Everything in Pro
• Claude Opus 4.6 (most powerful)
• 5× Pro usage limits
• Priority access
• Best for: heavy daily users

Max 20× $200 / month

• Everything in Max 5×
• 20× Pro usage limits
• For professionals running long agentic workflows
• Teams: $25–$150/seat

5 Minutes to Your First Session

Install & Setup

Two prerequisites, three commands, one browser login. That’s the full installation. Here’s each step with no assumed knowledge.

Prerequisite: Install Node.js

Claude Code needs Node.js (a JavaScript runtime) on your computer to run. Go to nodejs.org, download the LTS version for your operating system, and install it. That’s it — you don’t need to understand what Node.js is or does. It takes about two minutes.

Install Claude Code

# Step 1: Install Claude Code
$ npm install -g @anthropic-ai/claude-code
# Step 2: Launch it (this opens a browser login prompt)
$ claude
# Step 3: Verify — should print the version number
$ claude —version

After claude, your browser opens to Anthropic’s login page. Sign in with your account (the same one your Pro or Max subscription is on). After login, the terminal starts your first Claude Code session.

Navigate to a Folder You Want to Work In

Claude Code reads the folder you’re in when you launch it. Before starting a session, navigate to the relevant folder:

# Go to your Obsidian vault
$ cd ~/Documents/MyObsidianVault
$ claude
# Or your writing folder
$ cd ~/Documents/Writing
$ claude
# Or your desktop (for ad hoc tasks)
$ cd ~/Desktop
$ claude

Windows Note

Claude Code does not run natively on Windows. You need Windows Subsystem for Linux (WSL) — search “Enable WSL” in Windows settings, install Ubuntu from the Microsoft Store, then install Node.js and Claude Code inside WSL. Takes about 10 minutes the first time.

Start Here — Not With All the Features

Your First 5 Prompts

Don’t start by learning everything Claude Code can do. Start by doing five things that are immediately useful. These prompts work for anyone — no technical background, no setup beyond installation. Run one per day for your first week and you’ll understand Claude Code better than most people who’ve used it for months.

Each of these works in any folder with any kind of content. You don’t need Obsidian. You don’t need a specific file structure. You just need a folder with files and a Claude Code session pointed at it.

Free Download

Get the Creator Vault — Built for This

Pre-built templates, CLAUDE.md starters, AI prompt library, and the Obsidian CLI Quick Reference — everything organized and ready to use. Free in the Creator Vault.

Free. No spam. Unsubscribe any time.

The Memory System

CLAUDE.md — Your Persistent Memory Layer

Every AI tool has the same fundamental problem: it forgets you between sessions. You spend the first five minutes of every conversation re-explaining your preferences, your context, your constraints. CLAUDE.md solves this entirely.

CLAUDE.md is a plain Markdown file that Claude Code reads automatically at the start of every session. Whatever you put in it — your writing style, your project goals, things you want Claude to always or never do, context about your work — gets loaded into Claude’s memory before you type your first message. You brief it once, and every session starts already knowing you.

The Three Levels of Memory

# Level 1: Global — applies to every project
~/.claude/CLAUDE.md
  → Your writing voice, universal preferences, things Claude should never do
# Level 2: Project — applies to sessions in this folder
~/Documents/MyVault/CLAUDE.md
→ Current projects, folder structure, specific goals, local rules
# Level 3: Auto-memory — Claude’s own notes across sessions
~/.claude/projects/[project]/memory/
→ Decisions made, things Claude learned about your work, context it found useful

What to Actually Put in Your CLAUDE.md

The most valuable CLAUDE.md files are short, specific, and honest. Aim for under 100 lines. Here’s a template for non-developers:

# About Me
I'm [your name]. I write about [topic] for [audience].
My voice is [direct/warm/analytical]. I avoid [jargon/hype/passive voice].
# This Project
This folder contains [what’s here — vault, blog drafts, research notes].
My current focus is [current goal].
# Always Do

Match my writing voice when drafting — read existing files first
Ask before making significant changes to multiple files
Summarize what you did after completing a task

# Never Do

Don’t add ideas I didn’t have — refine, don’t invent
Don’t use exclamation points or marketing clichés
Don’t permanently delete anything without confirming first

The Command Layer

Slash Commands & Skills

In any Claude Code session, typing / opens a menu of commands. Some are built-in utilities. Some are skills — reusable instruction sets you create or install. Here are the ones worth knowing first.

Essential Built-In Commands

Skills — Your Custom Slash Commands

A skill is a reusable instruction set you write once and invoke with a slash command forever. Think of it as a macro or saved workflow. Skills live as Markdown files in a .claude/skills/ folder inside your project.

Three bundled skills ship with Claude Code:

Skills to Build for Writers & PKM Users

Read my Daily Notes from the past week. Summarize what I was focused on, surface anything unfinished, and suggest my top three priorities for today.

Using these notes as source material, draft a 400-word newsletter in my voice (see style guide in CLAUDE.md). Lead with the core insight. End with one actionable takeaway.

Scan my vault for notes related to [topic]. List the most relevant ones, suggest wiki-links I should add, and flag any gaps where a new note would strengthen the connection.

To create a skill: make a file at .claude/skills/morning-review.md, write the instruction set inside it (in plain English), and it immediately appears as /morning-review in your slash command menu.

Automation Without Code

Hooks — Make Claude Code Work Automatically

Hooks are “if this happens, then do this” rules for Claude Code. They fire automatically at specific moments — when Claude starts a session, when it edits a file, when it finishes a task — without you having to trigger them.

Think of hooks as the difference between a collaborator who just does what you ask and one who also handles the surrounding logistics automatically.

The Hook Events That Matter for Non-Developers

# This runs every time you open Claude Code in this folder
# It injects the current date and recent file changes as context
echo “Today is $(date ’+%A, %B %d, %Y’).”
echo “Recent changes:”
ls -lt —time-style=relative | head -5

Connect Everything

Set Up Claude Code with Obsidian

There’s no Claude Code plugin you install from inside Obsidian, and no in-app installer button — Claude Code is a terminal tool. “Setup” means pointing it at your vault folder from the command line (see Install & Setup above), then optionally connecting it to a running Obsidian instance for deeper access. Both levels, honestly laid out below.

MCP — Model Context Protocol — is the open standard that lets Claude Code plug into external tools. Without MCP, Claude Code reads files. With MCP, it can also interact with running applications, databases, calendars, browsers, and more. Think of MCP as a USB standard for AI integrations.

Claude Code + Obsidian: Two Ways to Connect

You don’t need MCP to use Claude Code with Obsidian. Your vault is already a folder of Markdown files — Claude Code can read and edit all of them immediately. But MCP unlocks deeper access: tag-based search, backlink resolution, graph data, and Obsidian’s full plugin ecosystem.

Without MCP (works immediately)

• Read and edit any note as a file
• Create new notes with your templates
• Search by content with plain English
• Reorganize folders and rename files
• Batch-process notes (summarize, tag, link)

With MCP (deeper integration)

• Search by Obsidian tag natively
• Resolve backlinks and graph relationships
• Use Obsidian templates and apply them
• Trigger Obsidian CLI commands
• Connect to Bases, Sync, and Publish APIs

Setting Up the Obsidian MCP

# Step 1: In Obsidian
# Community Plugins → Browse → "Claude Code MCP" → Install → Enable
# Step 2: In Claude Code — connect via MCP
$ claude
# Type: /mcp → add server → obsidian → http://localhost:22360/sse
# Step 3: Test the connection
> How many notes are in my vault? What are my most-used tags?
Your vault contains 2,847 notes. Top tags: #pkm (423), #project (187), #writing (91)…

Where It's Actually Strong

Non-Coding Tasks Claude Code Is Good At

Setup aside, here’s the honest list of what Claude Code does well for people who never touch code:

• Vault queries in plain English — “what did I write about X” across thousands of notes, no search syntax
• Drafting from your own material — newsletters, blog posts, and outlines built from your actual captures, not a blank page
• Style analysis — describing your own writing voice in specific terms you can paste into any AI tool
• Vault maintenance — finding orphaned notes, suggesting links, auditing content gaps across years of posts
• Research synthesis — reading dozens of documents at once and surfacing where they agree or contradict
• Repurposing — one piece of content reshaped into five formats in one pass

The eight workflows below walk through each of these with the exact prompts. For a broader look at non-coding use across professions, see Claude Code for non-coders.

Real Work, Real Results

8 Workflows for Creators & Knowledge Workers

These aren’t hypotheticals. Each one solves a specific friction point that comes up regularly in writing, PKM, and creative work. Start with the one that matches your biggest current frustration.

Morning Practice

Morning PKM Review

The problem: You open Obsidian each morning, scan your notes, and still feel unclear on what you should focus on. The context is there — you just can’t synthesize it fast enough.

Point Claude Code at your vault and say: “Read my Daily Notes from the past 7 days. Summarize what I was focused on, what feels unresolved, and what I keep returning to. Then suggest three priorities for today based on what you read.”

Claude reads your actual writing — not a summary you pre-made, but the real notes with all their ambiguity — and gives you a briefing that feels like talking to someone who read your journal. Do this three times and you’ll wonder how you started days without it.

Content Creation

Newsletter Draft from Scattered Notes

The problem: You have ideas captured all week — rough notes, half-sentences, saved links with annotations — but turning them into a coherent newsletter takes hours of painful synthesis.

Drop all your raw notes into a folder and say: “Read everything in this folder. Find the most interesting thread running through these captures. Draft an 800-word newsletter in my voice — lead with the insight, not the context. End with one thing readers can do today.”

Claude doesn’t write generic content — it synthesizes your specific captures. The draft will sound like you because it came from your actual thinking. Your job becomes editing, not starting from blank.

Voice & Style

Build a Personal Style Guide from Your Own Writing

The problem: You know what you want AI-written content to sound like, but you can’t describe it. Every “write in my voice” prompt produces something generic.

Point Claude at your best existing writing and say: “Analyze everything in this folder. Describe my writing style in specific, measurable terms: average sentence length, vocabulary level, what I emphasize, what I avoid, how I open and close pieces. Write this as a style guide I can paste into any AI tool.”

Once you have this style guide, paste it into your CLAUDE.md. Every future Claude Code session automatically matches your voice. The difference between generic AI writing and writing that sounds like you is this document.

Vault Maintenance

Orphan Note Audit & Link Suggestions

The problem: Your vault has hundreds of notes that captured good ideas but are disconnected — no links in, no links out. They sit in folders never to be found again.

Say: “Scan my vault for notes that have no links to or from other notes. For each one, read its content and suggest three existing notes it should link to, and why. Format the output as a checklist I can work through.”

This is the kind of task that would take you a weekend manually. Claude does it in minutes and often finds connections you genuinely missed — because it can read all 2,000+ notes simultaneously in a way your working memory can’t.

Content Strategy

Blog Content Gap Analysis

The problem: You’ve been publishing for years and have no clear picture of what you’ve covered, what’s missing, and what performs best — so you keep writing similar things.

Point Claude at your exported blog posts and say: “Read all of these posts. Map the topics I’ve covered, identify the gaps based on my audience and content themes, and suggest 10 post ideas I haven’t written but clearly should — based on patterns in my existing content.”

One creator reorganized 818 newsletter editions using this approach — Claude categorized, tagged, and identified the top-performing themes in under 30 minutes. What would have been a multi-day spreadsheet became a strategic content map.

Research

Research Synthesis Across Multiple Documents

The problem: You’ve been researching a topic for weeks across multiple documents, PDFs, and notes. You need a synthesis but can’t hold it all in your head at once.

Drop all your research into a folder and say: “Read all of these documents. Identify the key claims, find where they agree and contradict each other, and synthesize the most important insights into a structured outline. Flag anything that seems inconsistent or where I need more research.”

This is Claude Code operating as a research assistant with perfect recall across every document you give it. It doesn’t get tired, doesn’t miss things because it’s been reading for three hours, and doesn’t bring prior knowledge that contaminates your sources.

Repurposing

One Note → Five Content Formats

The problem: You write one good piece of content and then manually repurpose it five times — each format a separate effort that feels like starting over.

Say: “Read [filename]. Repurpose the core idea into: (1) a 280-character tweet thread opener, (2) a 150-word LinkedIn post, (3) a 5-bullet email newsletter teaser, (4) a YouTube video outline with 5 talking points, and (5) a one-paragraph blog intro. Match my voice for each format.”

This takes about 90 seconds and produces five usable drafts. Your editing job is minor. The formats are already adapted — not just reformatted text, but genuinely re-shaped for each context.

Weekly Review

Automated Weekly Review Generator

The problem: You know weekly reviews are valuable, but building one from scratch every Sunday is enough friction to skip it most weeks.

Say: “Read my Daily Notes from the past 7 days and all files I modified this week. Generate a weekly review note with: what I accomplished, what’s still open, the recurring theme across my captures, and three questions I should be asking heading into next week. Save it as ‘Weekly Review [date].md’ in my Reviews folder.”

This workflow becomes a skill (/weekly-review) that you run every Sunday. The review note exists before you’ve done any manual work — your job is to read it and add your own reflection, not to build it from nothing.

Learn This Before You Need To

5 Mistakes to Avoid

The same friction points trip up almost everyone. Knowing them in advance saves hours of confusion.

Mistake 1

Using it like Claude.ai — pasting content instead of pointing to files

The entire power of Claude Code is that it reads your actual files. Don’t paste content into the conversation. Say “Read the file at ~/Documents/draft.md” or navigate to the right folder first. File-first access gives Claude consistent, accurate context — not whatever you happened to paste.

Mistake 2

Mixing unrelated tasks in one session

Claude Code’s context window fills with the history of your session. Mix a newsletter task with a vault audit with a style analysis and the context gets noisy — Claude starts making inconsistent decisions. Use /clear between unrelated tasks. Treat each task like a fresh brief.

Mistake 3

Skipping CLAUDE.md entirely

Without CLAUDE.md, every session starts from zero. Claude doesn’t know your voice, your goals, your constraints. You’ll re-explain yourself every time and wonder why the results feel generic. Run /init in your project folder — it takes 10 seconds and transforms every future session in that folder.

Mistake 4

Vague prompts without structure or intent

“Help me with my writing” produces generic help. “Read draft.md, identify the weakest paragraph, and rewrite it to lead with the core claim more directly — keep my voice” produces specific, useful work. The CIF structure helps: Context (what is this), Intent (what I want), Format (how I want the output). Describe the outcome, not the process.

Mistake 5

Correcting Claude repeatedly in the same polluted session

When Claude gets something wrong, many people keep correcting it in the same session. After two failed corrections, the context is full of the wrong approach. Fix: after two attempts, use /clear and rewrite your original prompt incorporating what you learned from the failures. Don’t fight with a polluted session — reset and go again.

Go Deeper

Resources & Further Reading

Official

From This Site

Common Questions

FAQ

Do I need to be a developer or know how to code?

No. The workflows in this guide — morning review, newsletter drafting, vault organization, content repurposing, research synthesis — require zero coding knowledge. You write prompts in plain English and Claude Code acts on your files. The terminal is the only unfamiliar element, and opening it and typing a command takes about 30 seconds to learn.

Can I use Claude Code for personal knowledge management?

Yes — it’s one of the more common non-developer uses. Point Claude Code at your Obsidian vault (or any folder of notes) and use it to query your notes in plain English, draft from your own captures, run weekly reviews, and find orphaned notes. See the personal knowledge management section above, and the agentic PKM guide for the full framework.

Is Claude Code free?

Claude Code requires a paid plan. The Claude Pro plan ($20/month) is the entry point and includes Claude Code with Sonnet 4.6. The Max plans ($100–$200/month) unlock Opus 4.6 and higher usage limits. There is no free tier for Claude Code. For most non-developer use cases in this guide, Pro is sufficient.

What’s the difference between Claude Code and Claude.ai?

Claude.ai is a browser chatbot — you paste content in, it responds, you implement the result yourself. Claude Code is a terminal-based collaborator that reads your actual files, edits them directly, remembers your preferences via CLAUDE.md, and can handle multi-step tasks autonomously. Same underlying AI; fundamentally different interface and capability.

Is Claude Code safe to use with my personal files and notes?

Yes, with appropriate care. Claude Code reads and can edit files in the folder you run it from. Use /permissions to restrict what it can access if you have sensitive files. The default behavior is to ask before making significant changes. Start by pointing it at a non-critical folder until you’re comfortable with how it works. Nothing leaves your computer unless you use MCP to connect external services.

What is CLAUDE.md and why should I care?

CLAUDE.md is a plain text file that loads at the start of every Claude Code session as persistent memory. Without it, every session starts fresh and you re-explain your preferences every time. With it, Claude already knows your voice, your goals, what you want it to do and avoid, and the context of your project. Run /init in any folder to create one. It’s the single highest-ROI thing you can do in Claude Code.

Can Claude Code work with Obsidian?

Yes — in two ways. Basic: point Claude Code at your vault folder and it reads all your Markdown files directly. Advanced: install the obsidian-claude-code-mcp plugin to connect via WebSocket, giving Claude access to Obsidian’s search, tags, backlinks, and plugin ecosystem. The Obsidian CLI guide covers the full integration.

How do I set up Claude Code with Obsidian?

There’s no in-app installer. Install Claude Code once (npm install -g @anthropic-ai/claude-code), then cd into your Obsidian vault folder and run claude. That’s the entire basic setup — Claude Code now reads and edits every Markdown file in your vault. For deeper access (tags, backlinks, the graph), add the obsidian-claude-code-mcp plugin and connect it via /mcp. Full walkthrough in Set Up Claude Code with Obsidian above.

How do I install Claude Code in Obsidian?

You don’t — and no plugin makes that literally true. Claude Code runs in your terminal, not inside the Obsidian app. What you install is Claude Code itself (once, globally, via npm), then you run it from your vault folder so it can read and edit your notes. The optional MCP plugin connects the two more deeply, but Claude Code never becomes an Obsidian panel — it stays a separate tool pointed at the same files.

What model does Claude Code use?

Claude Sonnet 4.6 is the default model on Pro and the standard Max plan (released February 2026). Opus 4.6 — the most capable model — is available on Max 5× and Max 20× plans. Sonnet 4.6 handles all the workflows in this guide with excellent results. You’d only need Opus for very long documents, complex multi-step reasoning, or large agentic tasks across hundreds of files.

What are Claude Code skills and how do I make one?

A skill is a reusable instruction set you invoke with a slash command. Create a file at .claude/skills/skill-name.md inside your project folder, write the instruction in plain English inside it, and it immediately appears as /skill-name in your slash command menu. Three skills ship built-in: /simplify, /batch, and /debug. For non-developers, the most useful skills to build are around morning review, writing workflows, and vault maintenance.

What non-coding tasks is Claude Code good for?

Vault queries, newsletter and blog drafting from your own notes, style-guide extraction from your past writing, orphaned-note audits, research synthesis across many documents, and repurposing one piece of content into several formats. See the full list in Non-Coding Tasks Claude Code Is Good At above.