Nextretro MCP

Create and read retrospective boards, collect feedback, and manage action items from MCP-compatible AI assistants.

View on nextretro.io

Hosted MCP Server

npx add-mcp 'https://www.nextretro.io/api/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Let your AI agent prep your retro.

Your agent reads the sprint from Linear, Jira, GitHub and Slack and fills a Nextretro board with short, linked evidence cards. Your team runs the retro.

Works with Claude Cursor VS Code Antigravityand any MCP client

Your agent

Sprint 42 retro

Hidden

Went well

To improve

Only the facilitator sees these

Revealafter the facilitator has reviewed the cards

Sprint 42 retro

Revealed

Went well

Four planned issues closed in 1 to 4 days (PAY-101, PAY-104, PAY-107, PAY-115)

Refund API rollout got 9 celebration reactions (Slack, 1 thread)

To improve

Two issues slipped after 11 and 9 days in progress (PAY-110, PAY-112)

Three threads mention being blocked on staging (Slack, 3 threads)

From the sprint to the board, then back to the tickets

  1. Before the retro

    Ask your agent to prep it

    Say "Prep the retro for the sprint that closed on Friday." It reads Linear, Jira, GitHub and the public Slack channel you name.
  2. Before the retro

    It fills the board, hidden

    Three to five short cards per column, each ending in a link to its ticket or thread. The cards are hidden from everyone but you.
  3. In the retro

    You review, then reveal

    Edit or delete anything that is off, then switch the cards on. The team starts from evidence instead of from memory.
  4. After the retro

    It follows up on the actions

    Your agent checks each action item against its ticket and tells you what is done, what is open and what has no ticket.

Cards stay hidden

Nothing the agent writes reaches the team until you have reviewed it and revealed it.

Every card has a source

Each one ends with the ticket or thread it came from, so a claim can be checked in one click.

Private by design

No names, no quotes, public channels only. Nextretro never sends board content to an AI, and card authors and voters are never included.

Connect your agent

Add the Nextretro MCP server, then give your agent the skill. The skill is a plain file, so it works the same in every client.

Cursor

One click

Add to Cursor

Cursor asks you to approve the server, then opens the sign-in.

VS Code

One click

Add to VS Code

VS Code opens its install dialog. Needs a version with MCP support.

Claude Code

Copy and paste

claude mcp add --transport http nextretro https://www.nextretro.io/api/mcp

Run it in your terminal and sign in when it asks.

Claude app and claude.ai

Copy and paste

https://www.nextretro.io/api/mcp

Open Settings, Connectors, Add custom connector, paste the address and sign in.

Antigravity

Copy and paste

{
  "mcpServers": {
    "nextretro": {
      "serverUrl": "https://www.nextretro.io/api/mcp"
    }
  }
}

In the agent panel open MCP Servers, Manage, View raw config, and paste this in.

Any other client

Add a remote MCP server with this address. Clients without OAuth can use an API key from Settings.

Create an API key

Run it on a schedule

Use the ready-made Managed Agents configuration with Linear, Slack and Nextretro connected, and run it on the day each sprint closes. Add your uploaded skill id and one credential per server, then the board is waiting when the team opens the call. Then give any agent the skill file.

What your agent can do

Keys carry scopes, so you decide whether an agent can only read or also write. The server speaks the 2026-07-28 specification and still answers earlier clients.

  • list_board_types Read the available templates and their columns
  • create_board Create a board with seed cards, optionally hidden
  • add_card Add one more card to a board
  • get_board and summarize_board Read a board, with vote totals (no AI on our servers)
  • list_action_items See open and done actions
  • create_action_item and update_action_item_status Record a decision, mark an action done

Questions about agents

What does my agent need to connect?

The Nextretro MCP server at https://www.nextretro.io/api/mcp. Sign in with OAuth the first time, or create an API key under Settings, API keys. It works with Claude, Cursor, VS Code, Antigravity and any client that supports MCP over HTTP, including clients on the 2026-07-28 specification and earlier ones.

Does Nextretro send my cards to an AI?

No. Your agent runs in your own account with your own tools. Nextretro does not send board content to a language model, and board responses leave out card authors and voter identities.

What does it read from Slack?

Only public channels you point it at, and it records links and counts. The skill tells the agent not to quote messages, name people or read direct messages.

Why are the cards hidden?

So you can review what the agent wrote before the team sees it, and so pre-filled cards do not anchor what people write. You reveal them with the Hide cards switch on the board.

Does it cost anything?

The MCP server and the skill are free to use. Your AI provider may charge for running the agent.

Prefer to run the retro yourself? Start a free board