Claude Code in Zed: Two Ways to Set It Up (and Which I'd Pick)

Yes. Zed's own documentation lists two ways to run Claude Code-style work in the editor. The first is a Terminal Thread: a terminal Zed keeps in its Threads Sidebar, where you run the normal claude command and get the native Claude Code experience. The second is the Claude Agent, an External Agent you install from Zed's ACP Registry, which runs Claude inside Zed's Agent Panel through the Agent Client Protocol (ACP).

The thing to know up front: unlike the Claude Code VS Code extension and the Claude Code JetBrains plugin, there is no official Anthropic plugin for Zed. Anthropic's platform docs list the CLI, the desktop app, VS Code, JetBrains, the web and mobile. Both Zed paths are built and documented by Zed. That is fine, it just means you will not get things like the IDE diff viewer hookup that the official plugins add.

This is the simplest path, and it is the real thing. A Terminal Thread is a terminal that Zed files away as a thread, next to any agent threads you have open. Whatever you run inside it owns its own login, models, skills, instruction files and MCP setup. So when you run claude there, it is exactly the same Claude Code you would run in any terminal.

If you always want Claude Code in these threads, Zed has a setting for it. Set agent.terminal_init_command to "claude" in your Zed settings, or fill in the Terminal Thread Init Command field under AI in the Settings UI, and every new Terminal Thread starts Claude Code for you. You can open as many Terminal Threads as you like, each with its own entry in the sidebar, which makes running a couple of sessions side by side easy.

The second path puts Claude in Zed's Agent Panel as an External Agent. Open the registry with the zed: acp registry command, or go to Agent Settings, then External Agents, click Add Agent and choose Install from Registry. Pick Claude. It then shows up in the new-thread menu in the Agent Panel and the Threads Sidebar.

To pick how you pay, open a Claude Agent thread and run /login. Zed's docs say you can authenticate with an API key or with Claude Code where supported. One trap: an Anthropic API key you set up for Zed's own built-in agent does not carry over to the Claude Agent. They are separate.

What changes on this path is who owns what. Per Zed's docs, the External Agent usually owns its own models, login and tools, while Zed hosts the conversation. Zed's agent profiles and Zed Skills do not apply. Zed's MCP servers may be forwarded to the agent, and Claude's own files like CLAUDE.md may be read directly. In practice, expect your Claude setup to mostly carry over, but test it on your project before you rely on it.

My pick is the Terminal Thread. It is the real Claude Code with nothing in between, so every guide you read, every slash command and every setting works the same way. The ACP agent looks more at home in Zed, but it is another layer, and when something behaves differently you have to work out which side is responsible. If you like the chat look, try it, but keep the Terminal Thread as your fallback.

Either way, set your permissions deliberately. Claude Code permissions explained covers the modes, and since there is no IDE diff viewer hookup in Zed, approving edits as they come in is a sensible default while you get used to it.

My take: Zed users get a good Claude Code setup, it just comes from Zed rather than Anthropic. Because the Terminal Thread path runs the plain CLI, your settings, CLAUDE.md file and MCP servers move with you if you switch editors later. If you are setting up for the first time, Claude Code day one setup covers the first hour, and which Claude plan for Claude Code helps you pick the login.