Agent Deck, developed by Asheshgoplani, is an MCP server tool built with Go, designed for managing AI coding agent sessions. It functions as a TUI (Terminal User Interface) operating on top of a background daemon, which uses a per-user Unix socket for communication. This free tool provides a centralized platform for developers to control and monitor various AI agents like Claude Code, Gemini CLI, Aider, and Devin, facilitating the interaction with multiple AI development environments. Its primary purpose is to orchestrate and manage concurrent AI coding tasks efficiently.

What does Agent Deck connect to?

Agent Deck is a terminal session manager specifically for AI coding agents. It connects to and manages sessions for a variety of AI tools, including Claude Code, Gemini CLI, Aider, Codex, OpenCode, Pi, Devin, and Kiro CLI/IDE. The tool operates by observing their terminal sessions, providing real-time status updates, tracking tool calls, and monitoring prompts. Its internal communication relies on a per-user Unix socket between its TUI and background daemon.

Core features and tools

Agent Deck exposes its capabilities primarily through CLI commands and a TUI. Key commands include agent-deck session start, agent-deck session stop, agent-deck session fork, and agent-deck status. It also manages workspaces with agent-deck add --worktree, agent-deck worktree list, and agent-deck worktree cleanup. Additionally, it offers Sub-Agent Launch via helper scripts like "$SKILL_DIR/scripts/launch-subagent.sh" and first-class support for DeepSeek Harness (dsh) tools, enabling launch, resume, and session management.

Who is it for?

Agent Deck is designed for developers who frequently run and manage multiple AI coding agent sessions concurrently. It caters to those needing a unified interface to control and switch between agents across different projects. This includes users who want to explore various solutions by forking agent conversations or orchestrate complex workflows involving multiple specialist AI agents. Skip this if you primarily use only one AI coding agent at a time.

How to get started

No credentials are required for Agent Deck itself, as it primarily manages authentication for the AI agents it orchestrates. For its "conductor" mode, Slack user ID authorization is used.

Use cases

• Unified Session Management: Provides a single terminal UI to manage, group, search, restart, and switch between multiple AI coding agent sessions.
• Session Forking: Allows branching AI agent conversations to explore different solutions without losing original context.
• Agent Orchestration and Parallel Workflows: Enables setting up pipelines where an orchestrator coordinates tasks among specialist worker agents, supporting parallel execution.

Agent Deck by Asheshgoplani offers a centralized TUI for orchestrating diverse AI coding agents. Its Go-based architecture and Unix socket communication provide a focused environment for developers. The tool excels in managing concurrent sessions, facilitating agent orchestration, and enabling session forking for exploratory work. It acts as an important layer for developers engaged with multiple AI assistants. Try launching an "agent-deck session start" for a DeepSeek Harness session to experience its management capabilities.