If you mean Conductor (conductor.build): it’s a macOS app for running multiple Claude Code agents in parallel, where each agent gets an isolated workspace (separate copy + branch/worktree) so they don’t step on each other’s changes. (Conductor)
- Download + install Conductor on Mac. (Conductor)
- Add a repository; Conductor creates a first workspace (isolated copy + branch). (Conductor)
- Each new workspace is effectively “another Claude Code instance” working in isolation.
- Create more agents by creating more workspaces (e.g., ⌘N / new workspace flow depending on your setup). (Conductor)
Put a conductor.json in the repo/workspace root to automate setup and running:
{
"scripts": {
"setup": "npm install; ln -s \"$CONDUCTOR_ROOT_PATH/.env\" .env",
"run": "npm run dev"
},
"runScriptMode": "nonconcurrent"
}- setup runs on workspace creation (install deps, copy/symlink
.env, etc.). (Conductor) - run is the “Run” button (server/tests). (Conductor)
nonconcurrentis useful if you can’t run multiple dev servers simultaneously. (Conductor)
Create 4–6 workspaces and assign each Claude a single responsibility:
- PM/Planner agent: breaks down the task into PR-sized chunks + acceptance criteria.
- Implementer agent(s): each takes one chunk (API, UI, migrations, etc.).
- Test agent: writes/updates tests, runs them, fixes flaky parts.
- Reviewer agent: reviews diffs, flags risks, asks for refactors.
- Docs/Changelog agent: updates README, docs, release notes.
Then you (the human) do the “conducting”: review diffs, run tests, merge. Conductor’s workflow is explicitly built around “create workspace → develop → review/test → PR/merge → archive.” (Conductor)
Conductor supports MCP so Claude Code can access external tools/data sources. (Conductor) Example from their docs:
claude mcp add context7 -s user -- npx -y @upstash/context7-mcpConductor ships with its own bundled Claude Code (and Codex). (Conductor)
If you tell me what stack you’re on (Rails/Next/Django/etc.) and whether you want many parallel feature agents vs a tight reviewer/tester loop, I can suggest a concrete “workspace template” (prompts + role checklist + a conductor.json tuned for your repo).