Features

MCP Tools

TaskFlow exposes 50 tools across 10 domains via the Model Context Protocol:

Domain	Tools	Count
Tasks	create_task, list_tasks, get_task, update_task, update_task_status, delete_task, bulk_create_tasks, search_tasks	8
Projects	create_project, list_projects, get_project, update_project, delete_project, search_projects	6
Timers	start_timer, pause_timer, stop_timer, list_sessions	4
Activity	get_activity_log, log_debug, clear_activity_log, compact_activity_log	4
Analytics	get_analytics, get_timeline, get_tool_stats, get_task_cost	4
Notifications	list_notifications, mark_notification_read, mark_all_notifications_read, clear_notifications	4
Settings	get_setting, update_setting	2
Agent Comms	register_agent, list_agents, send_to_agent, ask_agent, check_messages, respond_to_message, broadcast_agents, check_broadcast, ask_user, check_response	10
Terminal	capture_terminal, send_keys	2
System	bootstrap, get_agent_instructions, create_checkpoint, get_checkpoint, list_checkpoints, clear_data	6

Real-Time Live Sync

Changes flow instantly from the MCP server to the desktop app:

1. AI agent calls an MCP tool (e.g., create_task)
2. MCP server writes to SQLite
3. SSE event fires on port 3456
4. Tauri app receives event and updates IndexedDB via Dexie.js
5. React UI re-renders

No polling. No delays. Sub-second updates.

Task Dependencies

Tasks can declare dependencies on other tasks. TaskFlow provides:

• Dependency graph: Interactive DAG visualization
• Cycle detection: Prevents circular dependencies
• Blocker awareness: Agents can query the graph to understand what's blocking them
• Auto-unblock: When a blocking task completes, dependent tasks can be unblocked

Multi-Agent Collaboration

Multiple AI agents can work on the same project simultaneously:

• Agent Registry: Agents register on startup and discover each other via list_agents
• Direct Messaging: send_to_agent for fire-and-forget updates, ask_agent for questions that need answers
• Broadcasts: broadcast_agents sends a message to all connected agents
• Task Assignment: Lead agent creates tasks and assigns work to other agents
• Shared Memory: log_debug with project_id creates entries visible to all agents

Remote Control

Optional relay server for controlling agents from anywhere:

• Outbound-only: Your local machine never accepts inbound connections
• Works behind NAT: No VPN or port forwarding needed
• Agent Inbox: Agents ask questions, you respond from any device
• Terminal Capture: See agent terminal output remotely
• Send Keys: Type into agent terminals from the remote app

Token Efficiency

MCP responses are optimized for AI context windows:

• 61% token reduction vs v1.0.0
• 72% smaller list_tasks payload
• Null fields suppressed from responses
• Compact JSON formatting