How to run multiple Claude Code agents at once (2026)

One Claude Code session is a good assistant. Three or four at once, each on its own task, is where the real speed comes from, if they don't trip over each other. This guide covers every way to run several Claude Code agents in parallel in October 2026, from plain terminal tabs to Anthropic's newer tools, and what each one costs your plan.

(I build Morphy, one of the apps that does this. It gets one short entry; everything else is Claude Code itself, from Anthropic's own documentation, checked on October 20, 2026.)

The one rule: give each agent its own copy of the files

Two agents editing the same folder overwrite each other's work. The fix is a git worktree: a separate folder with its own branch, sharing your repository's history. Every method below either uses worktrees for you or leaves it to you, so it's worth knowing what they don't keep apart:

Ports, databases and running processes are shared. Two agents starting a dev server on the same port still collide.
Gitignored files don't come along unless you list them in a .worktreeinclude file. That's how your .env gets into each copy.
Merge conflicts still happen, just later, when you bring the branches back together.

1. Terminal tabs with worktrees

The simplest way, and the one Anthropic's docs start with. In each terminal tab, start Claude Code with a name:

claude --worktree feature-auth

Claude Code creates a worktree under .claude/worktrees/feature-auth/ on a new branch, worktree-feature-auth, and starts the session there. Run it again with another name in a second tab, and you have two agents that can't touch each other's files.

Set it up once: add .claude/worktrees/ to your .gitignore, and list the gitignored files each copy needs (like .env) in .worktreeinclude. Dependencies need installing in each worktree; just ask Claude to.
Cleaning up: when you exit, Claude Code checks for unsaved work. A clean worktree is removed (a named session asks first); one with changes asks whether to keep it, and prints the command to resume it later.
Already in a session? Ask Claude to "work in a worktree" and it makes one.

Good for: two to four tasks you steer yourself. Watch out for: you switch between tabs to see who needs you.

2. Agent view: hand off tasks and check back

Agent view is one screen for all your background sessions. Run claude agents, type a task, and it starts as a row that shows what it's doing, whether it needs you, and when it's done. Peek at a row to reply without leaving the list, or attach to it for the full conversation. Sessions keep running after you close the screen.

Before a session from agent view edits anything, it moves into its own worktree, so you don't have to set that up yourself. You can also start one from your shell with claude --bg.

Good for: three or more independent tasks, like a bug fix, a pull request review and a flaky test, that you want to hand off. Watch out for: it's a research preview, so the screen and its shortcuts may change.

3. The Claude desktop app

The Code tab of the Claude desktop app runs sessions in parallel from a sidebar you can filter by status, each in an optional worktree, with a visual diff to review. It needs no terminal at all, and runs on Windows, macOS and Linux (beta).

Good for: working without the terminal, with review built in. Watch out for: at most two sessions side by side, and it needs a paid Claude plan. More in Claude Code desktop vs Morphy and the best Claude Code GUIs.

4. Subagents, inside one session

A subagent is a helper Claude starts for a side task: searching the code, running the tests, reading logs. It works in its own context and hands back a summary, so your main conversation stays clean. You can define your own (a reviewer, a test runner) and give each a cheaper model.

For one large change, like a rename across hundreds of files, the built-in /batch skill has Claude split the work into 5 to 30 subagents, each in its own worktree.

Good for: one task that breaks into pieces. Watch out for: every subagent draws on your plan too.

5. Agent teams: Claude hands out the work

With agent teams, one Claude session leads and starts teammates, each a separate Claude Code with its own context, sharing a task list and messaging each other. The lead plans, assigns and checks the work.

They're experimental and off by default; turn them on with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in your settings or environment.

Good for: a project Claude can split up and coordinate itself. Watch out for: teammates don't get worktrees, so give each one different files, and teammates in plan mode use about seven times the tokens of a normal session.

6. Dynamic workflows, for the very big jobs

A dynamic workflow is a script Claude writes that runs many subagents and checks their results against each other: a codebase-wide audit, a migration across 500 files, research that needs cross-checking. Claude Code tracks running workflows with /workflows.

Good for: jobs too big for a handful of subagents. Watch out for: it's a lot of agents, and a lot of usage.

7. Apps that run many agents for you

If you'd rather see every agent at once than switch between tabs, an app can do the bookkeeping:

Morphy (Windows, Mac, Linux): every chat is an agent on one canvas, so you see which one is working and which needs you. Claude Code runs next to Codex, Copilot, Gemini and others, any chat can work in its own copy of the project, and when one plan runs out, another AI can carry the conversation on. Free for 3 agents at a time.
Conductor (Mac only): a polished worktree-per-task app with review and pull requests. See Conductor vs Morphy and Conductor alternatives.
Orca and Emdash (Windows, Mac, Linux): free, open-source worktree apps.

All of them are compared in the best tools to run Claude Code and Codex in parallel.

How do you keep track of them?

Background sessions: claude agents shows every one and which need you.
Inside a session: /tasks lists what's running in the background, subagents included, and lets you check on or stop each.
Workflows: /workflows.
Sessions you run yourself can pass findings to each other with cross-session messaging.

What does it do to your plan?

Every session, subagent and teammate draws from the same Claude plan. Anthropic puts it plainly: running several sessions or subagents at once multiplies token usage. Five agents use a Pro plan about five times as fast, and the 5-hour and weekly limits apply to all of them together. See how to stop hitting Claude Code's usage limits, and consider giving some of the work to Codex on your ChatGPT plan.

Questions

How many Claude Code sessions can I run at once? There's no fixed number. Your plan's usage limits are the real ceiling, and they're shared across every session.

Can two sessions work on the same repository? Yes, as long as each has its own worktree. Without one, they edit the same files and overwrite each other.

Do worktrees share my settings? Mostly. Worktrees of one repository share its git history, project plugins and the permission approvals you give, and they read your project's skills, agents and commands from the main checkout when the worktree has none of its own.

Can I run Claude Code and Codex at the same time? Yes. They use separate plans, so it's also a way around limits. See Codex vs Claude Code.

References

Anthropic: Run agents in parallel, worktrees, agent view, subagents, agent teams, dynamic workflows, the desktop app, costs
Morphy: Chats and agents, including Its own copy