AI agents now have a persistent kanban board. The new @arvoretech/kanban-mcp gives agents structured task management that survives across sessions, with multi-chat coordination and semantic search via LanceDB.

What changed

• New @arvoretech/kanban-mcp package with 12 tools: list_boards, create_board, get_board, get_card, create_card, update_card, move_card, claim_card, release_card, search_cards, archive_card, delete_card
• Multi-session support — each chat identifies with a session_id, claims cards, and releases them when done. get_board shows active sessions with duration
• Subtasks — cards support parent_card_id for hierarchical task breakdown. get_card shows subtasks with status
• Semantic search via LanceDB with multilingual embeddings — find cards by meaning, not just keywords
• New mcp.kanban() helper in the RHM CLI for easy configuration

Multi-session coordination

When multiple chats are open in parallel, each one can claim cards to signal ownership. Other chats see who is working on what, preventing duplicate work. Cards track a full session log with timestamps.

Who is affected

Any workspace using RHM with AI agents. Add mcp.kanban() to your hub.config.ts and run hub generate.

Setup

import { defineConfig, mcp } from "@arvoretech/hub";

export default defineConfig({
  mcps: [
    mcp.kanban(),
  ],
});

npx @arvoretech/hub@0.14.0 generate