Meeting to Action Items: paste a transcript, get who-owns-what

★

Stop leaving meetings with a fuzzy memory and a long to-do list

The meeting ends, everyone nods, and within an hour half the room remembers the decisions differently. Two action items get dropped because nobody wrote down who owned them. The recap email you meant to send sits unwritten until Friday, by which point it's stale.

This kit does the writing. You drop in a transcript (paste it, or connect your notetaker's export), and Claude produces a tidy summary, the decisions the room actually made, and a list of action items with an owner and a date on each one. Then it drafts the follow-up email and the task entries, and waits for you. You read it, fix anything, and approve before a single message goes out.

Plain-English glossary, once: A transcript is just the text of who said what (your notetaker app makes one automatically, or you paste your own). Claude is the AI that reads it and does the summarizing and drafting. An action item is a single task with one owner and one due date, the thing that actually gets dropped when nobody writes it down. This kit's whole job is to never drop one.

Pairs naturally with two other lessons: Follow-Up Drafter (turn each action item into a sent-on-approval nudge) and Weekly Digest (roll a week of these recaps into one Monday summary).

1

Give it a brain: connect Claude (the setup wizard collects your key)

What's happening: The whole kit runs on one brain: Claude. It reads the raw transcript and turns "ok so Priya you'll own the deck, let's say by next Thursday" into a clean, structured action item with an owner and a date. The kit uses Anthropic's official toolkit and asks Claude for a structured recap (typed lists of decisions and action items), never loose prose, so the page can render it as a real table.

Go to console.anthropic.com and sign in.
Open API Keys → Create Key.
Copy the key (it starts with sk-ant-).

Your Claude API key goes here (the kit's setup wizard saves it on YOUR machine, not on this page)

Default brain: claude-opus-4-8 (sharpest). Lots of meetings? Set claude-haiku-4-5 to cut cost.

Soft CTA: the free kit runs the setup wizard for you with npm run setup, no config files to hand-edit. Grab the kit and run the wizard →

2

Get a transcript in: paste it, or connect your notetaker

What's happening: The kit needs the text of the meeting. The simplest path is to paste a transcript straight in. If you already use a notetaker (Granola, Otter, Fireflies, Zoom, Google Meet), point the kit at its export and it reads the file. Either way, the transcript stays on your machine: the kit only sends the text to Claude to summarize, and never stores it anywhere you didn't put it.

Where it comes from	How the kit reads it	Type
Paste / a `.txt` file	Drop text in, or point at a file path	built in
Granola / Otter / Fireflies	Export the transcript, the kit reads the export folder	export
Zoom / Google Meet	Save the auto-transcript, drop the file in `data/transcripts/`	export

Where your transcripts live (the kit watches this folder)

No notetaker yet? Pasting works perfectly. Start there.

3

Tell it what to pull out: summary, decisions, action items

What's happening: The kit hands Claude the transcript with a clear, plain set of instructions: write a short summary, list the decisions that were actually made, and extract every action item as a single task with one owner and one due date. If a date was vague ("next week") it converts it to a real date; if no owner was named, it flags the item as unassigned instead of guessing.

Why this keeps you honest: Claude is told, in plain rules, to pull owners and dates only from what was said. It does not invent who agreed to what. Anything ambiguous (no clear owner, a fuzzy "soon") lands in a clearly labeled needs-a-human list rather than getting a made-up name or deadline. You see exactly what was assumed before you approve.

↳ Roll these into a Monday digest: Weekly Digest →

4

Try it: paste a transcript, get a recap, send nothing

What's happening: Paste a meeting transcript (or tap a sample). The kit reads it, writes a summary, lists the decisions, and pulls out every action item with an owner and a date. Then it shows you the recap and the drafted follow-up email, and waits. This is confirm-before-send: it summarizes and drafts, you approve. This demo runs entirely in your browser. Nothing is sent anywhere. No API call, no email, no task created, no charge.

Meeting Notes · your workspace

Tap a sample transcript, or paste your own:

Paste the meeting transcript

Pure in-browser dry run: nothing is sent anywhere.

5

Approve → it emails the recap and creates the tasks

What's happening: Only after you tap Approve does the kit send the recap email to the attendees and create the task entries in your tool. Tap Edit to fix an owner, a date, or the wording; tap Decline and nothing leaves your machine. The draft in the demo above is wired to exactly this. Try the buttons: every one of them is a dry run.

Three promises the kit keeps:
• Nothing reaches an inbox or a task board without your explicit tap.
• The bot never invents an owner or a deadline it can't point to in the transcript.
• The transcript stays local. It summarizes and drafts; it does not upload your meeting anywhere you didn't choose.

In the kit, the recap renders to a clean email draft you approve, and the action items become draft task entries in whatever you use (the README shows the small add-ons for Todoist, Asana, Linear, or a plain Markdown file).

6

Bonus: one command turns a folder of transcripts into recaps

What's happening: When you fall behind, you don't re-process meetings one by one. Point the kit at your transcripts folder and npm run batch drafts a recap for each one. Every recap is still a draft you approve, and it still sends nothing until you say so. It just gets the writing done so your only job is the quick read-and-approve.

npm run dryrun     # summarize a transcript in your terminal, nothing sends
npm run recap      # draft the recap + action items for one meeting
npm run batch      # draft recaps for every transcript in the folder
npm start          # watch the folder and draft recaps as they land

Run it yourself: it's free and it's the real thing

The kit is a complete, runnable repo: a transcript in, Claude as the brain, a clean recap and a who-owns-what list out, plus a drafted follow-up email and task entries. MIT-licensed, runs on your machine, nothing crippled. Run npm run setup and you're recapping meetings.

Get the free kit on GitHub →

▶

Go live

With the kit downloaded and the checklist green:

npm install
npm run setup     # Claude key, transcript folder, who to email saved on YOUR machine
npm start

Drop a transcript into the folder (or paste one) and the kit drafts the recap and the action items right there; you approve before anything goes out. Prefer the command line? npm run dryrun summarizes from a text file. It also runs as an MCP server (npm run mcp) if you'd rather drive it from Claude, the same idea from the MCP lesson.

Want it wired to your notetaker, your task tool, and your inbox?

This kit is the DIY taste. We connect it to your real notetaker (Granola, Otter, Fireflies, Zoom), your live task tool (Asana, Linear, Todoist, ClickUp), and your inbox, so approved recaps email themselves and approved action items land as real tasks, and we run it for your whole team.

Work with us → ↳ Next: the Follow-Up Drafter