agtx: The Self-Managing Kanban Board That Automates Your Multi-Agent Workflow
If you're running multiple Claude Code sessions alongside other AI coding assistants, you know the pain: constant context switching, manual task delegation, and deciding what to work on next. A developer just released agtx—a terminal-native kanban board that manages itself via its own MCP server, automating your entire multi-agent workflow.
What It Does — Automated Agent Orchestration
agtx solves the parallel session bottleneck by creating a visual workflow where different AI agents handle different phases of development. You configure which agent handles which stage (e.g., Gemini for research, Claude for implementation, Codex for review), and the system handles the handoffs automatically.
The breakthrough is the orchestrator agent—a dedicated Claude instance that manages the entire board through its own Model Context Protocol (MCP) server. You add tasks to the backlog, press a single key, and the orchestrator triages, delegates, and advances tasks through your predefined workflow. The architecture follows this loop:
Orchestrator → MCP Server → DB → TUI → back to Orchestrator
You return to find pull requests ready for merge, with the system having managed the entire progression.
Setup — Terminal Installation with Plugin System
Installation is straightforward via the terminal:
git clone https://github.com/fynnfluegge/agtx
cd agtx
# Follow setup instructions in README
The system ships with a plugin architecture that lets you integrate spec-driven frameworks with a single TOML configuration file. You can plug in:
- GSD (Goal-Specific Development)
- Spec-kit
- OpenSpec
- BMAD (Business Model Analysis & Development)
Or define your own custom workflows. The TUI (Terminal User Interface) provides the kanban visualization while the backend handles all agent coordination.
When To Use It — Parallel Development and Complex Projects
Use agtx when:
- You're managing multiple concurrent features that require different AI strengths
- Your workflow has distinct phases (research, implementation, testing, review) that benefit from specialized agents
- You want to batch-process tasks and let the system handle prioritization and delegation
- You're working with spec-driven development and want to automate the translation from specifications to code
The orchestrator agent uses Claude's reasoning capabilities to make intelligent decisions about task assignment and progression, effectively creating a self-managing development pipeline.
The Technical Edge — MCP-Powered Automation
What makes agtx particularly powerful for Claude Code users is its use of MCP. The orchestrator agent communicates with the kanban system through a dedicated MCP server, which means:
- Standardized communication between Claude and the task management system
- Extensible architecture that can integrate with other MCP-compatible tools
- Direct agent control without manual intervention
This isn't just another task manager—it's a framework that turns Claude into a project manager for your other AI coding assistants.
Getting Started Today
Clone the repository and examine the example configurations. Start with a simple two-agent workflow (Claude for implementation, another agent for review) to see the automation in action. The TUI gives you real-time visibility into what each agent is working on, while the orchestrator handles the logistics.
For teams using multiple AI coding tools, agtx represents a significant step toward truly automated development workflows where human intervention is needed only for high-level direction and final review.
