🎥 Watch the full episode: How To 10X Your Notes: Obsidian + Claude AI Agents

📆 Published: 9th July 2025 🕒

Estimated Reading Time: 8 mins. Time saved: 53 mins! 🔥

Your note-taking app is about to become an AI-powered knowledge machine. Here's the step-by-step playbook for transforming Obsidian into an autonomous research assistant that works while you sleep.

Most people think Claude Code is just for writing software. They're missing the bigger picture. McKay Wrigley—developer, AI workflow expert, and creator of Takeoff's Claude Code course—has spent months extrapolating Claude Code's capabilities far beyond coding.

His insight? This isn't just an AI coding tool. It's a flexible agentic system that can automate knowledge work, research tasks, and note-taking workflows that previously required hours of manual effort.

In this comprehensive tutorial, Wrigley walks through 10 increasingly sophisticated workflows that transform Obsidian from a simple note-taking app into an AI-powered knowledge base. Whether you're a product leader managing research, a founder synthesising market intelligence, or anyone drowning in information, this guide demonstrates how to build autonomous systems that capture, organise, and synthesise knowledge whilst you're on the go.

This is the video that sparked my Claude Code journey and inspired me to write the Claude Code series. Ahead of tomorrow's number two in the series (Setting Up Your PM Automation Stack), I thought I'd share this to whet your appetite. It's a great podcast. I urge you to actually watch it after reading the summary.

The combination of Claude Code and Obsidian creates something fundamentally different from traditional note-taking. Obsidian stores everything as plain markdown files—text files you actually own, stored on your computer, accessible to any tool. Claude Code is an agentic AI system that excels at knowing when and how to use tools autonomously.

"The goal of this video is we are going to teach you how to build flexible agentic workflows that you can apply to all sorts of different tasks that you experience day-to-day," Wrigley explains. "This is why I like to refer to it as Claude Agent - we're going to extrapolate its coding capabilities so we can use it for anything."

The setup requirements are deliberately minimal. You need Obsidian (free download from obsidian.md) and Claude Code installed on your machine. Claude Code requires either a Claude subscription (starting at $20/month) or a Claude API key. The installation is straightforward (although I'm not going to lie, it does take a bit of a while to get used to terminal) - copy a single terminal command from the Anthropic docs and run it.

But here's what makes this powerful: because Obsidian uses plain markdown files, AI models understand them perfectly. There's no proprietary format to parse, no complex data structures. Just text files that Claude Code can read, edit, create, and organise with remarkable precision.

Wrigley demonstrates the basic setup by creating a fresh Obsidian vault, opening a terminal in that folder, and running the claude command. Within seconds, he's asking Claude Code what's in specific files, watching it search intelligently through the vault, and verifying it works. The entire setup takes less than 5 minutes.

Key Takeaways:

Before building any workflows, you need to establish the foundation: Claude Code's rules system. This is the most important configuration step, yet it's often overlooked.

Type /init in Claude Code and it creates a CLAUDE.md file at the root of your vault. This file becomes the system prompt injected into every single request you make with Claude Code. Think of it as the constitutional document that governs how Claude Code interacts with your entire knowledge base.

Wrigley emphasises this point repeatedly: "This file is going to have a significant impact on the quality of responses, how Claude actually works with your notes inside of your vault. You can make this as specific and personalised as you like."

The default rules file Claude Code generates includes guidance on file structure, formatting preferences, and basic interaction patterns. But the real power comes from customisation. Want Claude Code to never use tags? Add that rule. Prefer a specific date format? Specify it here. want UK English spelling? Add it. Have conventions for how you name files or organise folders? Document them.

One critical example Wrigley demonstrates: controlling tagging behaviour. By default, Claude Code might add tags to new notes. If you don't want tags, simply add "do not use tags" to your rules file. Instantly, every future interaction respects that preference without you having to specify it each time.

The rules file persists across sessions. Exit Claude Code, start a new session, and those rules are automatically loaded. This creates consistency in how your vault evolves, even as you work on different tasks weeks or months apart.

For advanced users, the rules file becomes the coordination layer for complex automations. You can reference other configuration files (like a dedicated tagging system file), set up folder organisation standards, and establish workflows that run consistently regardless of which specific task you're working on.

Key Takeaways:

Claude Code's interactive mode works like familiar AI chat interfaces, but (as you'll know if you've already read Part One in the Claude Code series,) with a crucial difference: it can actually modify files, create folders, and reorganise your knowledge base. Understanding the permission system and core commands prevents mistakes whilst unlocking powerful capabilities.

The basic interaction pattern is conversational. Wrigley demonstrates by asking Claude Code to pick a random topic (it chooses the Fibonacci sequence) and then requesting: "Please create a new note about that." Claude Code writes a comprehensive markdown file, complete with proper formatting, sections, and even tags at the bottom.

But here's where the permission system becomes important. Before creating the file, Claude Code asks for approval. You have three options: "Yes," "Yes and don't ask again this session," or "No and tell Claude what to do differently." This graduated permission model gives you control whilst reducing friction for repetitive tasks.

The file operations Claude Code can perform include:

Wrigley demonstrates the @ symbol for tagging specific files in your prompts. Type @, start typing a filename, and Claude Code autocompletes. This lets you give surgical instructions: "Can you add an interesting fact section at the bottom, but above the tags?" Claude Code edits exactly the file you specified, showing green lines for additions and red for deletions before you approve.

The /clear command resets your message history. This is essential when switching between tasks—you don't want Claude Code carrying context from a previous conversation into a new, unrelated workflow.

One powerful example: "Create a maths folder and put the related files in it." Even after clearing message history, Claude Code intelligently searches the vault, identifies the Fibonacci note as mathematics-related, creates the folder, and moves the file—all whilst asking for permission at each step.

Key Takeaways:

Right, speech-to-text again. Yes, I've covered this before. But stick with me, because Wrigley's approach here demonstrates why I keep coming back to this: it's the connective tissue that makes the entire Obsidian + Claude Code system sing. You're not just transcribing - you're thinking out loud whilst an AI system captures, structures, and organises everything in real-time.

Wrigley uses Wispr Flow (now called Flow), but emphasises the tool matters less than the principle. "You can speak tokens about four times faster than you can type them. If you're somebody like me who's basically working with AIs all day, every day, it's just much more efficient." (I've recently switched to Wispr Flow and it's absolutely brilliant, for example it automatically notices when you’re doing a list, so formats it as such, has a backtrack feature where you could be speaking away mid-sentence and say, "Oh no, actually change that," and it will go back and edit what you’ve said. Super handy.)

The workflow is simple: hold down a function key (on Mac) or use whatever trigger your speech-to-text tool provides, speak your prompt, and release. The text appears instantly in Claude Code, ready to send. No typing, no pausing to formulate sentences character by character.

The real power emerges with complex prompts. Wrigley demonstrates: "All right, Claude, I want you to create a new note. It should be your three favourite dinosaurs and why." Spoken in seconds, whereas typing would take considerably longer.

But the transformational use case is rambling ideation. "This is particularly useful when you just want to ramble for a while. I'll have these sessions where I'm rambling about a feature I want built for like three or four minutes. I can just hold my function key, I don't have to type, I don't have to really think through it. I can just dictate to the AI."

Claude Code's models are sophisticated enough to handle messy, jumbled speech. You don't need perfectly structured prompts. The natural, stream-of-consciousness style of speaking often produces better results because you're not self-editing as you type.

For Wrigley's coding workflows, this means describing features in natural language whilst walking around or thinking out loud. For research workflows, it means capturing sudden insights or research directions without breaking flow to hammer on a keyboard.

The tool Wrigley recommends (Flow) offers particular advantages: excellent Mac and iOS apps that sync seamlessly. Record voice notes on a walk using the mobile app, return to desktop, paste the transcription into Claude Code, and have it organise the thoughts into structured notes.

Key Takeaways:

This workflow is admittedly niche - using a code editor for note-taking - but demonstrates an important principle: because Obsidian stores plain text files you own, you can use any tool in your arsenal to work with them.

Cursor is a code editor with an exceptional autocomplete feature called Cursor Tab. Wrigley demonstrates opening an Obsidian markdown file directly in Cursor alongside his Claude Code terminal session.

Here's what makes this interesting: edit the file in Cursor, save it, and Obsidian instantly reflects the changes. It's the same file. You're not syncing between applications - you're editing the same plain text file from two different interfaces. this is the same when you ask Claude Code to edit the file. You can be working in the file and see Claude adding text to the file at the same time.

Cursor Tab's autocomplete becomes powerful for bulk editing. Change "three favourite dinosaurs" to "four" and Cursor Tab immediately suggests adding a fourth dinosaur. Hit tab to accept. Converting bullet points to numbered lists? Start the first conversion, and Cursor Tab detects the pattern and offers to complete the rest. Tab through each line without manual editing.

But here's where it gets clever: you can combine Cursor Tab with Claude Code commands. Wrigley demonstrates using Claude Code with the @ symbol to tag the file and give an instruction: "Please convert the bullet points to numbers." Claude Code processes the request, makes the edit, and because it's the same file, both Cursor and Obsidian show the updated version.

For some users, this creates a powerful workflow: open your entire Obsidian vault in Cursor, use Cursor Tab for rapid iteration on long files, run Claude Code in a Cursor terminal tab, and leverage both AI systems simultaneously on the same vault.

Wrigley acknowledges this is optional. "I know a lot of people are like, I don't want to pay for all these different products. No need to do that." But for those already using both Claude Code and Cursor for development work, applying these tools to note-taking unlocks unexpected efficiency.

For the record I have not yet seen the value in going into Cursor for this. I find Claude Code and Obsidian on their own are absolutely fine.

Key Takeaways:

Plan mode transforms Claude Code from an action-taking agent into a strategic planner. Press Shift+Tab twice to enable plan mode—Claude Code stops executing and starts planning instead.

The workflow Wrigley demonstrates reveals plan mode's power. Using speech-to-text, he gives Claude Code a complex research task: "I want you to go to the Anthropic docs and figure out how the Claude Code SDK works. I'm interested in building an Express server to build automations with the Claude Code SDK."

In plan mode, Claude Code doesn't immediately start writing files or searching documentation. Instead, it researches, thinks through the task, and presents a plan for approval. You see what it intends to do before it does anything.

During research, Claude Code demonstrates its web search capabilities—a bonus feature Wrigley highlights. It visits the Anthropic docs, searches for relevant SDK information, and compiles understanding without writing to your vault yet.

When the plan appears, you have two options: approve it or provide feedback to refine it. Wrigley chooses "No, keep planning" and adds: "I want this documented in a Claude Code SDK note." Claude Code revises the plan to include creating that specific note in the Obsidian vault.

Once approved, Claude Code executes the plan step-by-step. It creates an internal to-do system (visible in the terminal), tracks progress, and completes each action. The result: a comprehensive note documenting how to use the Claude Code SDK with an Express server, complete with code examples and explanations.

Plan mode becomes essential for multi-step workflows where you want oversight before execution. Research tasks that will create multiple files. Reorganisation projects that will move many notes. Complex analyses that require specific structure.

Wrigley emphasises the flexibility: "If you want to use plan mode for something that has like a dozen plus steps, totally can. You can just continue to iterate on the plan back and forth with Claude Code until it is configured to your liking."

Auto-accept mode can combine with plan mode. Approve the plan, let Claude Code auto-accept all the individual file operations, and watch it execute autonomously whilst maintaining strategic control through the planning phase.

Key Takeaways:

Custom commands transform repetitive workflows into single-slash shortcuts. This is where Claude Code evolves from a powerful tool into a personalised knowledge automation system.

Commands live in .claude/commands/ at the root of your vault. Each command is a markdown file containing a reusable prompt. The filename becomes the slash command.

Wrigley walks through creating a simple example: get-time.md. Inside, a heading and instruction: "Please get me the current time." Save the file, restart your Claude Code session (required for new commands to load), and type /get-time. Claude Code executes that prompt.

But the real power emerges with the $ARGUMENTS variable. This injects whatever you type after the slash command directly into the prompt.

Wrigley demonstrates with a daily-template.md command:

Now typing /daily-template I'm super blocked by the last few lessons of my Claude Code Obsidian workflow injects that context into the command. Claude Code creates a new daily note in the specified folder with those blockers documented.

The customisation possibilities are extensive:

Wrigley's personal vault includes commands for:

The pattern is powerful: identify workflows you repeat weekly or daily, codify them as commands, and reduce them to single slash prompts. Over time, your command library becomes a personalised automation suite tailored precisely to your knowledge work patterns.

Key Takeaways:

Knowledge base automation separates casual note-takers from power users. Claude Code can maintain your tagging system, create wiki-style links, and organise files autonomously—but only if you establish the rules first.

The foundation is a TAGS.md file at the root of your vault. Wrigley instructs Claude Code: "Please create a TAGS.md file at the root. In the file you need to define rules for our tagging system with hashtags inside of our Obsidian vault."

Claude Code generates a comprehensive tagging structure with categories, conventions, and rules. This becomes the source of truth. But here's the critical step: you must tell Claude Code to reference this file in the main rules file.

Using speech-to-text, Wrigley instructs: "Now we need to make sure to update our CLAUDE.md file. That's the rules for your system prompt. Anytime you go about tagging, you reference TAGS.md. This should be the source of truth for our tagging system. Make sure you also keep this updated over time."

Claude Code edits the CLAUDE.md file, adding a section that instructs it to always reference TAGS.md when tagging files. Now, in every future session, Claude Code automatically applies your tagging conventions without you specifying them.

The same pattern extends to wiki-style links and file organisation:

Wiki links: Create a WIKILINKS.md file defining when and how to create [[links]] between related notes. Claude Code automatically creates these connections as it works.

Organisation: Create an ORGANISATION.md file specifying folder structures, naming conventions, and categorisation rules. Claude Code organises new files correctly by default.

The power multiplies when combined with slash commands. Create a /tag-file command that analyses a specific note and adds appropriate tags based on your TAGS.md rules. Create a /tag-folder command that processes all files in a directory.

Wrigley demonstrates: "Say I wanted to create a command for tagging a file. Maybe in that command I say, 'Go inside whatever file I've tagged and update it with some tags.' You can start to chain commands with dynamic organisation."

As your vault grows to hundreds or thousands of notes, automated organisation becomes essential. Claude Code maintains consistency that would be impossible manually. Every new note gets tagged appropriately. Related concepts get linked. Files land in the correct folders.

And that really is a MASSIVE win.

Key Takeaways:

Most users run Claude Code tasks sequentially. Advanced users unlock parallel processing with sub-agents. Simply ask Claude Code to spin up multiple sub-agents, and it executes independent tasks simultaneously.

Wrigley demonstrates with a research task using speech-to-text: "I want you to spin three sub-agents up for a new research task. We are going to create a new note called AI Model Pricing. I want you to look up the pricing of OpenAI models, Claude Code models, and Google models. Put those in a table inside that note, and price everything per million tokens."

Claude Code creates three sub-agents immediately. Each one tackles a different pricing research task in parallel. The terminal shows all three tasks running simultaneously, each using web search to find current pricing information.

Instead of waiting for OpenAI pricing research to complete before starting Anthropic research, all three investigations happen at once. The time savings compound with task complexity.

Sub-agents excel at:

Wrigley highlights a practical pattern: slash commands for sub-agents. Create a /research command that automatically spins up sub-agents for common research workflows. Create a /write-sections command that drafts multiple parts of a document in parallel.

The completed example shows a comprehensive pricing table with OpenAI, Anthropic, and Google models, all priced per million tokens, with additional insights on prompt caching and batch API pricing. The research that might take 15-20 minutes sequentially completed in a fraction of that time.

For coding workflows, Wrigley uses sub-agents to research multiple documentation sources simultaneously. For note-taking, parallel processing works for gathering information from different MCP servers, processing multiple files, or researching various aspects of a complex topic.

Key Takeaways:

Model Context Protocol (MCP) servers transform Claude Code from a self-contained tool into an integration hub. MCP lets Claude Code connect to external services, APIs, and data sources, pulling information directly into your notes.

Wrigley demonstrates Context7, an MCP server optimised for software documentation, but emphasises the broader principle: MCP servers exist for Google Drive, Notion, Stripe, and countless other services. The pattern is consistent regardless of which server you're using.

Installing an MCP server is straightforward. Context7's documentation provides a command. In Claude Code, enable bash mode with ! and paste the installation command. Once installed, typing /mcp shows all connected servers.

The research workflow becomes powerful. Wrigley asks: "Use Context7 to find how to do generate object with Anthropic models inside of the Vercel SDK."

Claude Code uses the MCP server to:

No manual documentation browsing. No copying and pasting from multiple sources. Claude Code handles the entire research workflow autonomously through the MCP integration.

But the transformative examples go beyond documentation:

Google Drive integration: Pull files from Google Drive directly into your Obsidian vault. Wrigley: "Say I wanted to pull info from my Google Drive into my own personal vault. I could connect to the Google Drive MCP server, get that set up, and boom, I would be able to pull files."

Notion integration: Sync between Obsidian and Notion bidirectionally. Create polished Notion wikis from Obsidian research notes.

Stripe integration: For sales or customer success roles, query customer billing data before calls. "Maybe you're somebody running a business and you want to use the Stripe MCP to query customer data before a customer call."

Custom MCP servers: Wrigley reveals his personal workflow: "I have a custom-built MCP server where I can input a YouTube link, use a custom prompt command, and it will download the transcript of that YouTube link, take notes on it, and put it inside of my vault."

The pattern for any MCP server: find the server (search "[tool name] MCP"), install it with the provided command, restart Claude Code, and access it via /mcp. From there, the server's capabilities become available as tools Claude Code can use autonomously.

This is where the "agent" concept becomes clear. Claude Code isn't just a chatbot that generates text. It's an orchestration layer that knows when and how to use external tools to accomplish complex workflows.

Key Takeaways:

The final workflow demonstrates something remarkable: deploying Claude Code to the cloud so it works autonomously whilst you're away from your computer. This transforms Claude Code from an interactive tool into a background automation system.

The setup requires initialising your Obsidian vault as a Git repository and pushing it to GitHub as a private repo. Wrigley simply tells Claude Code: "I need you to initialise a new GitHub repo and push it to my GitHub as a private repository."

Claude Code autonomously:

Once the vault is on GitHub, type /install-github-app in Claude Code. This installs Anthropic's official GitHub integration, creating workflows that respond to GitHub issues and pull requests.

The autonomous workflow works like this:

Wrigley demonstrates from his phone, creating an issue titled "Test Integration" with the instruction: "Please make a new file in the vault in the /test folder called hello-world and write a funny joke for my viewers."

He adds a comment tagging Claude Code: "@claude-code please handle this. Thanks 😊"

The GitHub Action triggers immediately. Running entirely in the cloud, Claude Code creates the test folder, writes the hello-world file with a joke, commits the changes to a new branch, and updates the issue when complete.

Back on desktop, Wrigley asks Claude Code to merge the branch and pull the latest changes. The new file appears in his local vault with the joke intact: "Why don't scientists trust atoms? Because they make up everything."

The implications are significant. You can trigger research tasks whilst away from your computer. Create a GitHub issue with a complex research prompt before a meeting. By the time the meeting ends, Claude Code has completed the research, created organised notes, and pushed everything to a branch ready for review.

Wrigley acknowledges this workflow is the most advanced, but demonstrates the principle: "The goal is to get the wheels turning in your brain that you take the simplicity of what I just demonstrated and apply those concepts into practical real-world workflow examples."

For production use, you'd configure permissions, set up proper .gitignore files for sensitive data, and create dedicated commands for common background tasks. But the foundation is clear: Claude Code can work autonomously in the cloud, processing complex knowledge work whilst you're occupied elsewhere.

Key Takeaways:

McKay Wrigley's tutorial demonstrates that Claude Code is far more than a coding tool. It's a flexible agentic system that transforms how knowledge workers capture, organise, and synthesise information. The 10 workflows build progressively, from basic setup to cloud-deployed automation.

The immediate actions for anyone wanting to implement these workflows:

1 - Foundation:

2 - Automation:

3 - Integration:

4 - Advanced (Optional):

The compounding returns emerge over time. Each custom command you create saves time on every future instance of that workflow. Each MCP integration multiplies your research capabilities. The automated tagging and organisation system becomes more valuable as your vault grows to hundreds or thousands of notes.

Wrigley's core insight: "Claude Code is so much more than just a coding tool. It is a multi-purpose agent. It's very flexible. It's very malleable. You can get it to do whatever you want."

For product leaders, this means building a research system that captures competitive intelligence, synthesises market trends, and organises strategic thinking autonomously. For founders, it means offloading knowledge work to an AI system that works whilst you're focused on building. For anyone drowning in information, it means finally having a system that doesn't just store notes—it actively helps you think.

The note-taking app of the future isn't just a place to write. It's an AI-powered research assistant that knows your workflows, respects your preferences, and works 24/7 to build your knowledge base. That future is available today, and this guide shows you exactly how to build it.

I'll be taking this much further in my Claude Code series where I go through my personal setup and the roadblocks, hurdles, and challenges I faced and overcame in building it.

Final Key Takeaways:

Additional Resources:

Product Tapas Claude Code series #1 - Why every PM needs Claude Code (How I saved 200+ hours in 15 days using Claude Code for PM workflows)

Takeoff Claude Code Course: https://www.jointakeoff.com/ 23 lessons, 2 hours 22 minutes of comprehensive Claude Code training

McKay Wrigley's Blog: https://mckaywrigley.substack.com/ Accompanying article: "Claude Agent"

McKay Wrigley on X: @mckaywrigley

That’s a wrap.

I hope you're enjoying the new Claude Code series. Let me know!

Alastair 🍽️.