A mission control TUI for your fleet of AI agents

What it does

Agent Deck is a terminal UI that wraps AI coding agents—Claude, Gemini, OpenCode, Codex, and others—into a single session manager. It tracks whether each agent is running, waiting for input, idle, or in error, and lets you jump between contexts with keyboard shortcuts instead of tab-hunting. The tool also handles git worktrees, MCP server toggling, and per-group environment overrides so parallel agents on the same repo do not collide.

The interesting bit

The “conductor” feature turns one persistent agent into a supervisor that watches your entire fleet via Telegram, Slack, or Discord, escalating stuck sessions to your phone while answering routine questions itself. There is also an MCP socket pool that shares MCP server processes across sessions via Unix sockets, which the README claims cuts memory use by 85–90% and reconnects automatically after a crash.

Key highlights

• Fork any Claude conversation instantly to try alternate approaches without losing history.
• Toggle MCP servers per-project or globally from inside the TUI; the tool restarts the agent automatically.
• Status-aware fuzzy search with filters like @ for waiting and $ for error states.
• Built-in git worktree lifecycle: spin up isolated branches for parallel agents, then merge and clean up from the same interface.
• Per-group and per-conductor overrides for CLAUDE_CONFIG_DIR and environment files, so work and personal Claude accounts stay separated.

Verdict

If you run multiple AI agents daily and feel like your terminal is becoming a zoo, this is worth a look. Developers who only keep one Claude session open at a time will find it overkill.