Get started

Choose the harness where you want Muster to run. The deterministic CLI is shared, but installation, parallelism, isolation, and enforcement belong to the active harness.

Native CLI laneClaude CodeTen slash-command modes, plugin hooks, native agents, and native worktrees.Install for Claude Code →Native CLI laneCodexTen skill modes, scoped profiles, advisory hooks, and receipted worktrees.Install for Codex →Native CLI laneKimiTen namespaced skills, native subagents, permission rules, and explicit Init limits.Install for Kimi →Local MCP laneCoworkDeterministic MCP tools with a verified sequential path and proof before parallel use.Connect Cowork →Private/local · proof-gatedChatGPT WorkConditional developer lane through Secure MCP Tunnel. Requires a separately billed OpenAI Platform API key; a configured connection alone is not proof of native invocation.Review Work requirements →

Cost and service boundary

Muster's CLI makes no model calls. Claude Code, Codex, Kimi, and Cowork use the account or subscription of their active native harness and do not require a Muster-hosted runtime. ChatGPT Work is the documented exception: its private/local transport uses Secure MCP Tunnel and a separately billed OpenAI Platform API key, distinct from a ChatGPT subscription.

Already installed? Use the task-to-mode quickstart. For side-by-side capability details, see Harness support.