--- title: "Run multiple Claude Code agents in parallel | microManager" description: "Run several Claude Code agents at once, each in its own git worktree, with the Claude Code you have installed and signed in. Free, with diffs and PRs." url: "https://micromanager.win/agents/claude-code" updated: "2026-11-08" --- # Claude Code in microManager microManager runs the Claude Code you already have installed and signed in, through the Claude Agent SDK, so every chat is a full Claude Code session with your CLAUDE.md, settings, hooks and MCP servers, and the repository’s own once you trust it. You can run as many as you like at once, each in its own git worktree. Updated 8 November 2026 ![A microManager window while Claude Code, on Haiku, works in the workspace Reject overlong account names: edit cards showing its changes to test/accounts.test.ts and test/app.test.ts as diffs, a Run the test suite card that was allowed once, and its reply still streaming above a Working… line. The sidebar lists two more ledger-api workspaces, and the Changes panel lists three changed files.](https://micromanager.win/screenshots/workspace-claude-chat.png) Claude Code in the middle of a turn: each edit shows as a diff, the test run it asked for was allowed once, and its reply is still streaming. Codex and Grok each work in another workspace. ## How do I run several Claude Code agents at once? Create a workspace for each task (**⌘N**, or Ctrl+N on Windows and Linux) and pick Claude Code. Each workspace is its own git worktree on its own branch, so the sessions never edit the same files, and each can have more chat tabs (**⌘T**) for a second Claude working beside the first. The sidebar shows which sessions are working, which need an answer and which are done, and a notification tells you when one finishes. See [parallel agents](https://micromanager.win/features/parallel-agents). ## How does microManager run Claude Code? Each chat tab is a Claude Code session started through the Claude Agent SDK, with the SDK pointed at **your** `claude` executable: the one on your login shell’s `PATH`, or a wrapper script you choose. The session runs in the workspace’s worktree with the workspace’s environment, loads your user settings, and the repository’s project and local settings once you [trust the repository](https://micromanager.win/docs/scripts#why-did-a-script-from-the-repository-not-run), and keeps Claude Code’s own system prompt. microManager appends a short note about the workspace: its name and folder, the repository’s main checkout, the branch and its target branch, the shared `.context/` folder, the reserved ports and its own tools, then your general prompt from Settings → Prompts. ## Which models and effort levels can I use? The model picker offers `opus`, `sonnet`, `haiku` and `fable`, and accepts any model id you type, so new models work the day your Claude Code supports them. Effort goes from low through medium, high and xhigh to max. Both can change per chat. ## How do permissions and questions show up? Pick a permission mode per chat: **Ask first**, **Accept edits**, **Plan**, **Auto** (Claude Code’s own reviewer model decides each request) or **Bypass permissions**. New chats start in the `permissions.defaultMode` of your Claude Code settings, or in the mode you set for Claude Code in Settings → Agents. When Claude Code asks to use a tool, a card shows what it wants to do, with **Allow**, **Deny** and, when Claude Code suggests one, an always-allow choice such as **Allow all edits this session**. At the end of planning, **Approve plan** or **Keep planning** answers its plan. When it uses its `AskUserQuestion` tool, you get a card with its options (single or multiple choice, or your own text), and your answer goes straight back to it. ## What else does a Claude Code chat show? Streamed replies with formatted code, its thinking (collapsed), a card per tool call (commands with output, edits as diffs, searches, subagents), its to-do list while it works, and each turn’s duration, tokens and cost. **Esc** stops a turn; checkpoints before every message let you roll a turn back, and the session rewinds with it when your Claude Code version supports that. ## Can I use a wrapper script instead of claude? Yes. Set the Claude Code profile’s command to any executable that behaves like `claude`, such as a script that sets a proxy or chooses an account. If Conductor was set up with a custom Claude Code path, the import uses it. ## What tools does Claude get from microManager? The `micromanager` MCP server runs inside the session, so Claude can call `GetWorkspaceDiff`, `DiffComment`, `GetDiffComments`, `GetTerminalOutput` and `GetWorkspaceInfo`. For questions it uses its own built-in `AskUserQuestion`. See [MCP tools](https://micromanager.win/docs/mcp-tools). ## What do I need? Claude Code installed and signed in, so that `claude` works in your terminal. On Windows, Claude Code runs its shell commands in Git Bash when Git for Windows is installed, and in PowerShell otherwise. Then add a repository, press **⌘N** (Ctrl+N on Windows and Linux) and pick Claude Code. ## Where to go next - [Agents and profiles](https://micromanager.win/docs/agents) - [Codex in microManager](https://micromanager.win/agents/codex) - [Run agents in parallel](https://micromanager.win/features/parallel-agents) ## Frequently asked questions ### Do I need an Anthropic API key for microManager? No. microManager starts your claude CLI, which uses whatever sign-in it already has. microManager itself has no account and never sees your credentials. ### Will my CLAUDE.md, hooks and MCP servers apply? Yes. Sessions always load your own Claude Code settings. The repository’s project and local settings (its CLAUDE.md, .claude hooks and permission rules, and .mcp.json) load once you trust the repository, which repositories imported from Conductor already are; then everything behaves as it does in your terminal. ### Can I continue a Claude Code session I started in Conductor? Yes. Importing from Conductor brings each chat over with its Claude session id, loads its history from Claude Code's own transcript file, and the next message continues the same session.