taskboard

Taskboard includes a built-in MCP server that exposes 22 AI-friendly management tools for working with projects, tickets, teams, and board data via Model Context Protocol. Once the MCP server is running, you can connect an AI assistant (e.g., Claude, Claude Desktop) or any MCP-enabled client to issue commands such as listing projects, creating tickets, moving tickets between statuses, and querying or updating teams. The MCP interface uses JSON-RPC over standard I/O, so the server can be consumed from terminal tools or desktop integrations, enabling AI-driven planning, automatic task breakdowns, and board synchronization with your assistant. Your Taskboard data is stored locally in SQLite and accessible to all connected MCP clients, including the embedded web UI.

To use, start the MCP server via the Taskboard binary with the mcp argument (as shown by the example configurations). From there, connect your MCP client and begin issuing the 22 available tools (Projects, Teams, Tickets, Board, Subtasks) such as list_projects, get_project, create_ticket, move_ticket, get_board, create_subtask, and batch_create_subtasks. The tools are designed to let an AI outline work, create tickets from high-level prompts, and manage progress directly within your Taskboard instance.

Prerequisites:

• Go 1.24+ and Node.js 22+ (as noted in the README).
• A macOS/Linux environment suitable for building Go binaries.

Installation options:

Option 1: Homebrew (recommended for macOS)

brew tap tcarac/taskboard
brew install taskboard

Option 2: From source

git clone https://github.com/tcarac/taskboard.git
cd taskboard
make build

Notes:

• The build produces a single binary named taskboard that includes the server, CLI, and embedded frontend.
• After building or installing, the MCP server can be started using taskboard mcp (see usage for more details).
• The project requires Go 1.24+ and Node.js 22+ for development; the runtime binary does not require a separate runtime environment beyond the binary itself.

Tips and notes:

• The MCP server uses JSON-RPC over stdio. When connecting via Claude Desktop or other MCP clients, point the client to the Taskboard binary with the command and arguments (e.g., command: /path/to/taskboard, args: [mcp]).
• Data is stored in SQLite locally. On first run, migrations run automatically. You can customize the database path with the --db flag on commands that support it.
• The MCP API exposes 22 tools organized under Projects, Teams, Tickets, Board, and Subtasks. Use these to orchestrate planning and execution with AI assistance.
• The web UI and CLI share the same underlying data; actions performed through MCP will reflect on the Kanban board immediately.
• If you encounter port or API issues, ensure the MCP server is started and listening as part of the Taskboard process (default UI is on port 3010).