Run a multiplayer coding session with each person's Claude Code

Claude Squad lets each participant connect their own Claude Code as an agent, with fresh context per feature and spec and review passes.

What They Did

The creator built Claude Squad as a self-hosted shared session where each participant runs one CLI command to connect their local Claude Code as an agent. In chat, an orchestrator agent splits the requested feature into parallel tasks, each Claude Code works on its own branch, merges are applied in sequence into one PR, and people step in when flagged merge conflicts need manual resolution.

How It Went

It runs on free-tier Vercel, Supabase, and Partykit, keeps API keys on each person’s machine, and splits token cost across participants. The author says task decomposition is still the main bottleneck; at the time of the thread, conflict handling and post-merge review were still early and lacked much real-world data.

Worth Knowing

Each feature was built with a fresh Claude Code context plus spec and review passes, and collaborators join a session by running a single npx connect command.