AI Card Assistant

Overview

Every card on your canvas has a built-in AI assistant. Click the sparkles iconin the card header to open the chat panel. The assistant understands your card's context — tasks, priorities, assignees, connected cards, and notes — so you can manage everything through conversation.

The chat is real-time and collaborative. All workspace members who have the chat open can see messages as they arrive. Each person's messages appear in their own color.

What You Can Do

Multi-Step Execution

The assistant can handle complex requests that require multiple actions. It decides the best strategy automatically:

• Sequential:"Create a 7-day content plan" — creates tasks one by one, each with the right details.
• Batch:"Set story points for all tasks" — updates multiple tasks in a single turn.

Each action shows a live indicator as it runs, so you can follow along.

How It Works

1. 1

   Open the chat

   Click the sparkles icon in any card's header.

2. 2

   Type your request

   Write in natural language — the assistant understands your card's context.

3. 3

   Watch it work

   The assistant executes actions in real-time. Tool indicators show what's happening.

4. 4

   See the results

   Changes appear on the canvas immediately thanks to real-time sync.

Real-Time Collaboration

Chat history is persisted and shared across the team. When multiple members have the same chat open:

• Messages appear for everyone in real-time
• Online avatars in the header show who's currently viewing
• Each person's messages are colored with their unique team color
• The AI addresses people by name and understands multi-user context

Message History

Chat messages are saved in Firestore and persist across sessions. When you reopen a chat, previous messages load automatically. Older messages can be loaded by clicking Load earlier messages at the top.

Use the clear button (broom icon) in the header to delete the entire conversation history for a card. This requires confirmation.

External AI Agents (MCP)

AI Card Assistant isn't limited to manual chat. You can connect external AI agents — like Claude Code, Cursor, or Windsurf — directly to your cards using the Model Context Protocol (MCP).

When an agent subscribes to a card, it starts listening in real-time. Any message you type in the card chat is instantly forwarded to the agent. The agent can respond, create tasks, update priorities, and take actions — all through the same chat interface.

How it works

• After running npx @isteam/mcp@latest setup, the agent shows up in your Team Members panel as a draggable badge with a 6-character name you chose (e.g. HOME01, DEV001).
• Drag the badge onto any card to attach the agent. The card gets an MCP subscription glow and the agent starts listening in real-time. Each card can hold at most one agent at a time.
• You type a message in the card chat as usual. The agent receives it automatically — no extra setup, no webhooks to configure.
• The agent responds via the chat. You see "Claude is thinking..." while it processes, and the response appears as a regular message.
• The agent can call any available tool: create tasks, update fields, move items, add comments, log time — everything the built-in assistant can do.
• Remove the badge from a card to detach — the agent goes idle (stays running), ready to drag onto a different card. Running in background daemon mode lets the agent auto-execute any task assigned to its card without a terminal open.

Use cases