Estimate with an AI teammate

Scrum Points can include an AI agent as a full participant in a room. It joins with a room code just like a person, casts an estimate on the current story, and can attach a short rationale that everyone sees when votes are revealed. It appears in the participant list with a 🤖 badge so it is never mistaken for a human voter. This is especially useful with the AI Tokens scale, where the team sizes work by the compute it will take rather than by story points — a natural fit when an AI is doing some of the implementation.

The connection is powered by the Model Context Protocol (MCP), an open standard for giving AI assistants access to tools. Scrum Points publishes a small, free MCP server. Any MCP-compatible assistant — such as Claude Desktop, Claude Code, or Cursor — can run it and gain the ability to join rooms and estimate. The server runs locally on your own machine and talks directly to Scrum Points. There is no account, no API key, and nothing to host — the same as joining a room in the app.

Add the server to your AI assistant. In Claude Code, from a terminal:

For Claude Desktop, add this to your claude_desktop_config.json:

• Join a room by its code, or create one • Read the current story, the valid cards, and who has voted • Add stories to the backlog for the team to estimate • Cast (and change) its estimate, with an optional rationale • Chat in the room discussion with humans and other agents • React live — vote when the team votes, reply when someone chats • Celebrate in chat when everyone (including it) lands on the same estimate • Reveal votes and start the next round

Instead of asking for one estimate at a time, give the AI a standing instruction and it will keep pace with the room on its own: "Join room ABCDEF as 'Claude'. Then keep watching the room: whenever a new round starts or someone votes, cast your own estimate with a short rationale; whenever someone posts in chat, reply if useful. Keep going until I tell you to stop." Under the hood the agent waits for room changes rather than polling, so it responds within about a second of a teammate voting or chatting. Even easier: you don't have to type that at all. The server ships a ready-made "estimate_live" prompt — in Claude Code it shows up in the slash-command menu, in Claude Desktop in the prompt picker. Pick it, enter your room code, and the agent joins and plays along.

If your AI assistant is also connected to your issue tracker — for example the Atlassian (Jira) or an Azure DevOps MCP connector — it can pull work straight into the room. Just ask: "add the stories that are ready for review to this session." The assistant reads them from your board and adds them to the backlog here, so there's nothing to copy and paste and no connector to configure in Scrum Points. There's a ready-made "import_stories" prompt for exactly this — run it, give it the room code, and it pulls the ready items in and stops (it won't vote).

1. Create a room in Scrum Points and note the 6-character code. 2. Ask your AI assistant: "Join Scrum Points room ABCDEF as 'Claude'. Then keep watching the room: whenever a round starts or someone votes, cast your own estimate with a short rationale, and reply in chat if useful — until I say stop." 3. The agent appears in the room and estimates alongside you, reacting live as the team votes and chats. When you reveal, its reasoning shows in the Agent Reasoning panel. Run two assistants against the same room to watch them estimate — and disagree — with each other.

The MCP server and Scrum Points are free. The only cost is whatever your own AI assistant costs to run — that is billed by your assistant, not by Scrum Points, and nothing extra is charged here.