Retro Rabbit

Blog

MCP for Retrospectives: Seven Ways to Use Claude Code or Cursor With Your Retros

Connect Claude Code or Cursor to your retros over MCP: seven workflows for prep, notes, themes and action items, plus the guardrails to agree on first.

Retro Rabbit TeamUpdated 8 min read
On this page (5 sections)
  1. What an MCP server adds to a retro tool
  2. Seven retro workflows for your AI tool
  3. Guardrails to agree on first
  4. How to connect Claude Code or Cursor
  5. Related

Most of the work that comes out of a retro happens somewhere else: in your editor, your terminal, or a chat with an AI assistant. Until recently, getting retro context into those places meant copying notes out of a board and pasting them in. MCP takes that step away. Your AI tool asks the retro tool directly, as you, and gets the real board back.

We built an MCP server into Retro Rabbit, so the examples below use it. Here's what an MCP server can and can't do for a retro, seven workflows with prompts you can copy, the two-minute setup, and the guardrails to agree on with your team before anyone connects.

What an MCP server adds to a retro tool

MCP, the Model Context Protocol, is an open standard for connecting AI tools to outside services. A service publishes a set of tools, and the assistant decides when to call them. You ask a question in plain words, and the assistant picks the calls it needs.

Retro Rabbit's server has eight tools. Six read, two write.

ToolWhat it doesSomething to ask
list_teamsThe teams you're on"Which Retro Rabbit teams am I on?"
list_roomsA team's rooms and their current retros"Which rooms does my team have?"
list_retrosOne room's retros, with their action items"List the retros in the Platform room."
get_notesA retro's board: columns, notes, votes, groups and action items"What did people write in the Sprint 14 retro?"
my_action_itemsYour action items across the team's rooms"Which action items do I still owe?"
room_action_itemsEvery action item in one room, by retro"Show every action item in the Platform room."
add_noteAdds a note to a column of a retro, as you"Add a note under Didn't go well: checkout tests timed out again."
add_action_itemAdds an action item to a retro"Add an action item: give staging an owner and an alert."

Just as useful is what it can't do. It can't vote, edit or delete notes, mark an action item done, start a retro, or send an action to Jira or Linear. Those still happen in the retro, with people in the room.

Seven retro workflows for your AI tool

Each one starts with a question you'd ask anyway. The difference is that the assistant answers from the board instead of from your memory.

1. Catch up before the retro

Five minutes before the meeting is when people try to remember what was agreed last time. Ask instead.

Find the last retro in the Platform room. What were the top three
themes by votes, and which action items from it are still open?
Quote notes exactly, and say if something isn't on the board.

The quoting rule matters. A summary you can't trace back to the notes is a summary you can't check. The same goes for the checks in our retro prompts guide, which still apply when the assistant fetches the notes itself.

2. Add notes while it's fresh

Most retro notes are written on the day of the retro, from memory, so the board tilts toward whatever happened that week. If you live in an editor, add a note the moment something goes wrong, without switching windows.

Add a note to the current retro in the Platform room, under
"Didn't go well": the release build failed twice on the
flaky payment test.

The note is added as you, and it follows your own show-name setting. Keep notes about systems and work, never about a person. A note typed in the heat of the moment is still a note your teammates will read.

3. Check what you owe

Run this before standup or at the start of your week. It works across every room on the team.

Which retro action items do I still owe? Group them by room,
oldest first, and flag any that are older than two sprints.

4. Start on an action inside your codebase

This is where an AI coding tool does more than a chat window can. Pull the action item into the session, then let the assistant help with the work itself.

Read my open retro action items and find the one about flaky
tests. Then look through this repo's test suite for tests that
use fixed sleeps or real network calls, and list the worst five.

The action item stays in Retro Rabbit, and its owner still reports back at the next retro. The assistant just shortens the gap between agreeing to do something and starting it.

5. Find the problems that keep coming back

A problem raised once is noise. One raised in four retros in a row is the thing to fix, and it's hard to spot when each retro is a separate board.

Read the last four retros in the Platform room. Which problems
come up in more than one? For each, list the retros it appears
in and quote one note from each. Don't merge problems that only
sound similar.

Bring the answer to the retro as a question, not a verdict. The team decides whether it's really the same problem.

6. Draft the sprint recap

Stakeholders want two paragraphs, not a board. Let the assistant draft them from the real notes.

Draft a short recap of the Sprint 14 retro for our engineering
manager: the top themes by votes, the action items with owners,
and last retro's actions marked done or still open. Don't name
anyone except action owners.

Read it before you send it. It's a draft, and you're the one putting your name on it.

7. Audit a room's actions before you plan

Before sprint planning, look at follow-through across the whole room rather than one retro.

Show every action item in the Platform room from the last three
retros. Which have no owner, and which have been open for more
than a sprint? Keep it to a table.

Unowned actions are the ones that quietly die. If you track them in a tracker, our Jira and Linear guide covers the setup that keeps them visible.

Guardrails to agree on first

An AI tool connected to your retros acts as you, with your access. That's what makes it useful, and it's why a few rules belong in your working agreements before anyone connects one.

👁️

It sees what you see

A key reaches only the teams its owner is on, with the same checks as the web app. Notes posted without a name come back without an author, except your own.

✍️

Approve the writes

Claude Code and Cursor can ask you before each tool call. Keep that on for adding notes and action items, at least at first.

📤

Know where it goes

What the assistant reads goes to that AI tool under its own terms. Check your company's AI policy, and tell your team which tools are connected.

🔑

One key per tool

Create a separate key for each tool or computer, so you can revoke one without breaking the rest. The Integrations page shows when each key was last used.

Two more habits help. Keep owner picks in the room: the assistant can add an action, but the team should agree who owns it. And never paste an assistant's theme list into the retro as fact. It's a starting point for the conversation, not the conversation.

How to connect Claude Code or Cursor

Setup takes about two minutes. The MCP server page has the full details.

Connect an AI tool to Retro Rabbit

  1. 1

    Create a key

    Sign in to Retro Rabbit, open Integrations from the dashboard sidebar, and under AI tools name a key and choose Create key. It starts with rr_ and is shown once, so copy it then. You need a verified email, and guests can't create keys.

  2. 2

    Add the server to your tool

    Point the tool at https://api.retrorabbit.io/mcp with an Authorization: Bearer header carrying your key. The exact setup for Claude Code and Cursor is below.

  3. 3

    Ask something

    Start small: "Which Retro Rabbit teams am I on?" If that works, try the first workflow above.

For Claude Code, run:

claude mcp add --transport http retrorabbit \
  https://api.retrorabbit.io/mcp \
  --header "Authorization: Bearer rr_your_key"

For Cursor, add this to your MCP settings file and set RETRORABBIT_API_KEY in your environment:

{
  "mcpServers": {
    "retrorabbit": {
      "url": "https://api.retrorabbit.io/mcp",
      "headers": {
        "Authorization": "Bearer ${env:RETRORABBIT_API_KEY}"
      }
    }
  }
}

A few limits, so nothing surprises you. The server uses API keys, not OAuth, so connectors in the Claude and ChatGPT apps can't use it yet. It answers when asked and doesn't push updates. Keys are personal: there are no team-wide keys. It's free on every plan, including the free plan for one team, and you can see the rest of the integrations on their own page.

Frequently asked questions

What is an MCP server for retrospectives?

MCP, the Model Context Protocol, is an open standard that lets an AI tool call outside services. A retro tool's MCP server lets an assistant such as Claude Code or Cursor read your retro boards and action items, and add to them, without copy and paste.

Which AI tools can connect to Retro Rabbit's MCP server?

Tools that can connect to a remote MCP server over HTTP and send an Authorization header, such as Claude Code and Cursor. Tools that can only connect through an OAuth sign-in, such as connectors in the Claude and ChatGPT apps, aren't supported yet.

Can the AI see who wrote each note?

It sees a board the way you do. Notes posted without a name come back without an author, except your own. Notes it adds for you follow your own show-name setting, just like notes you add in the browser.

Can it change or delete retro notes?

No. It can read boards and action items, add a note and add an action item, all as you. It can't vote, edit or delete notes, complete action items, start a retro, or send an action to Jira or Linear.

Where does our retro data go?

To the AI tool you connected, under that tool's own terms. Retro Rabbit answers the tool's requests and doesn't push data anywhere. Check your company's AI policy before you connect a work tool, and tell your team.

Does the MCP server cost extra?

No. It's on every plan, including the free plan for one team. You need a verified email to create a key, and guests can't create one.

How do I disconnect an AI tool?

Revoke its key on the Integrations page. It stops working right away. The page also shows when each key was last used, so you can spot one you forgot about.

Run retros that actually lead to change

Your team adds notes during the sprint, CarrotIQ suggests groups and names them, and unfinished items carry into the next retro. Pro adds CarrotIQ summaries and suggested action items.

Free for one team · No credit card