Does Claude Code need its own API key for this?
No. runnrr starts the Claude Code CLI you already have installed, under the login you already pay for. There is no separate runnrr agent key.
How to run Claude Code from a Kanban card
How to run Claude Code from a Kanban card
Put the task on a board, attach the note and files it depends on, then press play. Claude Code starts on your Mac, reads the card first, and waits for your approval on every tool call.
Short answer
Here is the short version.
In runnrr you run Claude Code from a Kanban card by pressing play on the card. runnrr starts Claude Code as a local process under your own CLI login, hands it the card title, description, attached markdown context, and linked files, then waits for you to approve each tool call before it acts.
The card is the scope. Because the context is already attached, you are not pasting a project briefing into a fresh chat every time, and the run history stays on the card instead of scattered across terminal sessions.
Step by step
Run Claude Code from a Kanban card in three steps
Step 1
Get the task onto a card
Create the card by hand, scan it out of a Granola call, push it from Claude Code over MCP, or capture a selection anywhere on your Mac with the capture shortcut. What matters is that the task exists as a card and not as a line in a chat you will lose.
Step 2
Attach the context the agent will need
Link the markdown note, meeting transcript, decision, or files the task depends on. This is the single step that decides whether the run is useful: Claude Code reads these before it starts, so anything you leave out is something it will guess at.
Step 3
Press play, then approve each step
runnrr spawns Claude Code locally under your own login. It reads the card, then asks before every tool call. Approve, reject, or stop the run outright. When the work is done the card carries the run history and moves on to review.
What you end up with
- A scope the agent cannot drift out of, because the card is the boundary.
- No re-pasting: the meeting note and linked files travel with the task.
- A run history attached to the card rather than lost in terminal scrollback.
- Local execution under your own Claude Code login, with no second subscription.
FAQ
Questions people ask before trying it
What exactly does Claude Code see when I press play?
The card title and description, the markdown context saved on the card, and the files you linked to it. It reads those before it takes its first action.
Can I stop a run halfway through?
Yes. Every tool call waits for approval, so you can reject a single step or stop the whole run at any point.
Does the run go through runnrr servers?
No. Claude Code runs as a local process on your Mac reading local files, and no part of the run reaches our servers. Other runnrr features can leave your machine when you opt into them, such as cloud sync, the Granola scan, and public note links. The privacy page covers each.
Who this is not for
- Teams that need shared assignment, sprints, or issue tracking. runnrr is a personal board.
- Anyone wanting a fully autonomous agent. Every tool call stops for your approval by design.
- Windows or Linux users. The board and local agent runs are macOS only today.
Start now
Stop re-explaining your work to your AI
runnrr turns the follow-ups into cards your AI can finish. Free. No sign-up.
Download for MacmacOS 13+ · Web version coming soon
Related runnrr pages