asheshgoplani/agent-deck

What it solves

Agent Deck provides a centralized "command center" for managing multiple AI coding agents (such as Claude Code, OpenCode, Pi, and Codex) simultaneously. It eliminates the need to juggle multiple terminal windows or sessions by providing a single interface to track, switch between, and orchestrate a fleet of agents working across different projects.

How it works

It uses a Terminal User Interface (TUI) and a web UI to manage agent sessions. The system organizes agents into groups and allows users to switch between them with a single keystroke. It integrates with tmux for status notifications and supports advanced session management features like forking conversations, attaching Model Context Protocol (MCP) servers, and managing Claude skills.

Who it’s for

Developers who use multiple AI coding agents across various projects and need a way to organize their workflows, track costs, and manage agent state without manual configuration of config files.

Highlights

• Session Forking: Instantly create branches of conversations to explore different technical approaches without losing context.
• Conductor Orchestration: Set up a "conductor" agent that supervises other worker agents and can be notified via Telegram, Slack, or Discord.
• MCP & Skills Management: Toggle MCP servers and Claude skills per project or globally through a dedicated manager without editing config files.
• Git Worktrees: Support for isolated working directories and branches per agent to prevent conflicts when multiple agents work on the same repository.
• Socket Pooling: Reduces MCP memory usage by 85-90% by sharing MCP processes across sessions via Unix sockets.
• Cost Tracking: Includes a built-in cost dashboard to monitor spending across sessions.
• Smart Status Detection: Real-time polling to identify if an agent is running, waiting for input, idle, or in an error state.