How I run many Claude Code agents at once without losing track

Description and use

Switchboard centralizes management of multiple Claude Code agents, providing a needs‑you queue, fleet overview, Jira project import, automatic account failover, device sync via Tailscale, and process cleanup.

Problem

Managing multiple Claude Code agents simultaneously leads to idle time, lost context, lack of overview, dependency confusion, usage limits, device fragmentation, and resource leakage.

Outcome

Switchboard reduces idle agent time, provides a unified view of agent fleet, automates dependency sequencing, handles usage limits, enables cross‑device access, and cleans up resources, improving developer productivity.

Workflow

• 1. Install Switchboard locally. 2. Configure multiple AI engine logins. 3. Import a Jira epic as a Switchboard project. 4. Launch agents; each starts in its own worktree. 5. Agents report turn status via built‑in MCP tool. 6. Switchboard aggregates waiting turns into the Needs‑you queue. 7. Use the board view to monitor agent states. 8. Approve or reject agent‑generated notes. 9. Let Switchboard auto‑switch accounts on usage limits. 10. Pair devices with Tailscale for remote access. 11. Monitor and stop orphaned processes after idle period.

Evidence

With two agents you notice. With eight, an agent asks "Should I also update the mobile client?" at 10:14, and you find it at 11:30, and then you're like que?

Source

Everything waiting on me goes into one list: Needs you.

Source

I made a module that is importing the Jira epic as a project.

Source

Switchboard holds several logins per AI engine.

Source

I came up with the idea of pairing computers and phones over Tailscale, with no relay server.

Source

I added tracking the processes each agent starts, even after they detach, and showing them per agent with Stop and Keep.

Source