A paper-craft retro board with a first-class MCP server and CLI. Your agents read and write the same board your humans do, under the same auth and audit trail. Skip to § 4 to wire it up.
Post-its, columns, votes, a discussion phase, action items. The only new behaviour is the seam between the board and the rest of your work.
A board has three required properties: columns, a phase, and a membership. Phases run brainstorm → vote → discuss → close in that order. Notes are anonymous by phase, then attributed once discussion opens.
RetroSync's coach reads every board your team produces — then surfaces the patterns the room can't see across its own history.
By the third sprint every team is telling a story it hasn't realised it's telling. The same word in three Mad columns. The same person owning every unfinished action. A Glad count that quietly halved between April and May.
The AI watches the corpus, not the meeting. It runs once per retro and surfaces what the room can't see across its own history — so the next sprint starts with the signal, not the noise.
One personal access token, two transports. Every editor and every script reads and writes the board you're on right now.
RetroSync exposes the same board over three surfaces: a UI for humans (§ 2), an MCP server for AI editors (§ 4.1), and a retrosync CLI for terminals and CI (§ 4.2).
Tokens are scoped — read for watch-and-summarise, read,write for full teammate behaviour — and revocable per laptop, per CI job, per agent.
We stopped writing retros for the calendar and started writing them for the work. By Tuesday, last sprint's board is half-shipped.
A column-set, a phase order, a default voting budget. Open to remix.
Mint a token in 30 seconds. No credit card.