How I Automatically Extract Action Items from Meetings Using Claude Code
By Meg Faibisch Kuhn

I built an automation that extracts action items from my meeting notes and adds them to my task tracker. Every time I finish a meeting, a script runs automatically, pulls out the actual tasks, and queues them up for my review.
I'm not technical. But Claude Code made this possible.
Watch me explain this on TikTok →
What This Automation Does
Here's the workflow:
- I have a meeting and take notes in Granola (new users get 100% off their first month with my link)
- After the meeting ends, a script automatically runs
- The script reads my meeting notes and extracts actual action items—things I need to DO, not just discussion points
- Each task gets added to my dashboard in a "pending" state
- I review each item and either approve or deny it
- Approved tasks become real tasks with full context
The key is that last part: I still have a manual review step. The AI doesn't just dump tasks into my list. It queues them up, and I decide what actually makes it through.
What's great is that since it has all the meeting context, each task comes with a description that includes the relevant details—who's involved, why it matters, what was discussed. I don't have to remember anything. I just reference the task when I'm ready to work on it.
Why I Built It This Way
The automation matches my personal branding and website design. Same colors, same feel. This isn't vanity—it's about reducing cognitive friction. When my tools look consistent, my brain processes them faster.
I also wanted conservative extraction. The AI is trained to only pull out ACTUAL tasks—things that start with action verbs like "send," "schedule," "create," "review." It ignores discussion points, context, and anything that isn't a clear action item.
Fewer high-quality tasks is better than a noisy list of everything mentioned in a meeting.
The Prompt I Used
Here's the prompt I used to set this up with Claude Code. Feel free to grab it and adapt it for your own system:
I want to create an automated workflow that extracts action items from my Granola meeting notes and adds them to a task tracker.
Setup:
- I use Granola for meeting notes (I have the MCP integration connected)
- I want action items to go into [my dashboard / Notion / a JSON file / etc.]
- I want to review items before they become real tasks (pending → approved flow)
Requirements:
1. Only extract ACTUAL tasks (things I need to DO), not discussion points or context
2. Tasks must start with action verbs (send, schedule, create, review, test, prepare, cancel, etc.)
3. Be conservative - fewer high-quality items is better than many noisy ones
4. Each task needs:
- Title: concise action item starting with a verb
- Description: 1-2 sentences of context (why it matters, who's involved, relevant details from the meeting)
5. Match tasks to existing projects when possible based on keywords
6. Store a link back to the original Granola meeting notes
7. Run daily on a schedule (or on-demand)
My existing projects are: [list your projects with keywords, e.g., "Website rebuild (Acme Co)", "Monthly reporting (Northwind)"]
Can you help me set this up?
How to Adapt This for Yourself
A few things you'll need to customize:
Where your tasks go: I built mine into a custom dashboard on my website. You might want tasks in Notion, a JSON file, Todoist, or somewhere else. Just tell Claude Code where you want them.
Your project list: If you want tasks automatically matched to projects, include your project names and any keywords that would identify them (client names, project codes, etc.).
The trigger: Mine runs automatically after meetings. You could also run it manually, on a daily schedule, or however makes sense for your workflow.
The review flow: I have a pending → approved flow because I don't want AI dumping tasks directly into my list. You might trust the extraction more and skip the review step.
What You'll Need
- Claude Code (obviously)
- Granola for meeting notes (new users get 100% off their first month with my link)
- The Granola MCP integration connected to Claude Code
- A place to store your tasks
The Bigger Picture
This is what I mean when I talk about building your own operating system with AI. It's not about any single automation—it's about creating a personalized infrastructure that works the way your brain works.
My task extraction looks like my website because it IS part of my website. My meeting notes flow into my task list because that's how I think about work. Everything is connected because I connected it.
You don't need to be technical to build this. You just need to be clear about what you want and willing to iterate with Claude Code until you get there.
Want help setting up automations like this? Book a Claude Code session and we'll build something together.
Build Your Own Personal OS
Loved the idea of a custom task system? My Personal OS guide walks you through building a private command center for your tasks, projects, and finances — managed through plain English conversation with Claude Code.