AgentBoard

CLI task management for AI agents. A kanban board that agents can use programmatically.

📋 AgentBoard — 2 projects, 5 tasks

  brandfetch (3 tasks)
  ├─ 🔴 P0 [ab-001] Deploy API              kai    🔄 IN_PROGRESS  2h ago
  │  └─ 🔵 P2 [ab-002] Write migrations      kai    📋 TODO
  └─ 🟡 P1 [ab-003] Add pricing page         —      📋 TODO

  cleanprompts (2 tasks)
  └─ 🟡 P1 [ab-004] Bias detection pillar    —      📋 TODO

Why

AI agents (Claude Code, Codex, OpenClaw) need a way to coordinate work. AgentBoard gives them a shared task board — SQLite-backed, zero config, CLI-first.

• No server — single SQLite file at ~/.agentboard/board.db
• No setup — npx @andupetcu/agentboard add "Do the thing" and you're running
• MCP support — 15 tools exposed via stdio for Claude Code / Codex integration
• Sub-50ms — every operation is fast

Install

npm install -g @andupetcu/agentboard

Or use directly:

npx @andupetcu/agentboard

Quick Start

# Create a task
agentboard add "Build REST API" --project myapp --priority p0 --assign claude

# View the board
agentboard board

# Start working
agentboard start ab-001

# Add progress notes
agentboard comment ab-001 "Endpoints done, writing tests"

# Mark complete
agentboard done ab-001 "All endpoints + tests passing"

Commands

Tasks

| Command | Description | |---------|-------------| | add <title> | Create a task (--project, --priority, --assign, --parent, --tags, --due) | | list | List tasks (--status, --project, --assigned, --stale, --tree, --json) | | show <id> | Full task details + event log | | update <id> | Update fields (--status, --priority, --title, --assign) | | start <id> | Set status to in_progress | | done <id> [summary] | Mark complete | | block <id> <reason> | Block with reason | | unblock <id> | Resume blocked task | | cancel <id> [reason] | Cancel task | | comment <id> <text> | Add a progress note |

Views

| Command | Description | |---------|-------------| | board | Kanban view grouped by project (--project to filter) | | summary | Quick stats (total, by status, by project) | | stale | Tasks with no updates past their threshold | | timeline <id> | Full event log for a task |

Agents

| Command | Description | |---------|-------------| | agent register <id> | Register an agent (--type, --capabilities) | | agent heartbeat <id> | Update last_seen timestamp | | agent list | List all agents | | agent health | Status check (active / idle / dead) |

MCP Server

AgentBoard includes an MCP server for AI agent integration:

agentboard-mcp

This exposes all 15 operations as MCP tools over stdio. Add to your Claude Code or Codex MCP config:

{
  "mcpServers": {
    "agentboard": {
      "command": "agentboard-mcp"
    }
  }
}

Available Tools

task_add, task_list, task_show, task_update, task_start, task_done, task_block, task_unblock, task_cancel, task_comment, board, summary, agent_register, agent_heartbeat, agent_health

Task IDs

Sequential, human-readable: ab-001, ab-002, etc. Customize the prefix with AGENTBOARD_PREFIX env var.

Priorities

| Level | Label | Icon | |-------|-------|------| | p0 | Critical | 🔴 | | p1 | High | 🟡 | | p2 | Medium (default) | 🔵 | | p3 | Low | ⚪ |

Status Flow

todo → in_progress → done
     → blocked → in_progress
     → cancelled

Configuration

| Env Var | Default | Description | |---------|---------|-------------| | AGENTBOARD_DB | ~/.agentboard/board.db | Database path | | AGENTBOARD_PREFIX | ab | Task ID prefix |

Stack

TypeScript, SQLite (better-sqlite3), Commander, Chalk. ~2,000 lines, 59 tests.

License

MIT