Skip to content

These pages document master, which is unreleased and in development. The Quick Start installs the latest stable release; anything newer than that tag is marked in the text.

Anthropic Agent Teams (experimental, advanced option)

Claude Code v2.1.32+ has an experimental feature called Agent Teams that coordinates multiple Claude Code sessions on the same machine via a shared task list and mailbox at ~/.claude/teams/. Disabled by default; enable with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1.

Use only when all of these apply:

  • The task genuinely benefits from cross-teammate dialogue (e.g., competing-hypotheses debugging, parallel reviewers comparing notes), not just parallel execution.
  • Each teammate’s work is independent and heavy enough that summarizing back to a lead would lose important context.
  • You can afford significantly higher token cost — approximately 7× when teammates run in plan mode (per Anthropic’s cost docs). Actual cost depends on teammate count and mode.
  • You don’t need /resume or /rewind mid-team — neither restores in-process teammates.

Don’t use for routine multi-file feature builds. The single-session subagent pattern in agent-team-protocol is cheaper, recoverable, and produces equivalent results for the common case.

Limitations: one team per session, lead is fixed for the team’s lifetime, no nested teams, split panes don’t work in VS Code’s integrated terminal / Windows Terminal / Ghostty.