Tool runs multiple Claude Desktop accounts on one Mac
An unofficial MIT-licensed bash script gives each Claude account its own window and sign-in while all of them read the same Claude Code configuration.
What shipped, and when
A small open source project called claude-multi-desktop, published on GitHub on 2026-11-16, is built around one promise: running multiple Claude Desktop accounts at the same time on a single Mac. As of 2026-11-22 the repository showed 300 stars and 19 forks, collected in the seven days since it was created, according to the project's README. It carries an MIT license and lists HTML as its main language.
The problem it targets is narrow. Claude Desktop holds one sign-in at a time, and opening it again brings the same window forward, so someone with a personal account and a work account has to sign out to switch. The tool's author describes it as built for exactly that case: one person, two accounts of their own that use different email addresses.
Separate sign-ins, one Claude Code setup
Each launcher is a small app that opens the Claude.app already installed on the machine using Electron's public --user-data-dir flag. The README says it makes no copy of Claude.app, re-signs nothing and runs nothing in the background. Sign-ins live in per-account folders under ~/Library/Application Support/Claude-NAME, so each window keeps its own chats, projects and plan.
The shared part is Claude Code. Every account reads the same ~/.claude, which holds skills, CLAUDE.md, memory, MCP servers, hooks and settings. Write a skill once and each account loads it the next time it starts a session. That is the design choice that separates this from tools that keep accounts fully isolated, Claude Code setup included.
There is a second consequence. All accounts record their running Claude Code sessions in the same place, so a session in one account can see and message sessions in another through Claude Code's ListAgents and SendMessage tools. A claude-multi-desktop peers command prints which session runs in which account, with an account name, a status and a session label. The README's own example is asking the trip planning session in a personal account to check a hotel list, with the message reaching that session without copy and paste.
Installing it, and the first-launch warning
Homebrew is the shortest route: brew install jackson-cmd/tap/claude-multi-desktop, then claude-multi-desktop add Work to create a launcher and claude-multi-desktop open Work to sign in with the second account. A launcher for the account already in use comes from add Main --default, because once another window is open, the normal Claude icon only brings that window forward. Launchers land in ~/Applications and can take a custom colour, a coloured letter icon, a letter badge on the Claude icon, or a picture.
The README also documents a .pkg installer, a .dmg with a double-click app, npm, PyPI and a Claude Code plugin. Neither the installer nor the disk image is signed with an Apple Developer ID, so macOS blocks the first launch; the README walks through the Privacy & Security steps for macOS 15 and later, and the right-click Open route for macOS 14 and earlier. Because every account runs the same Claude.app, one update covers all of them, but the README says to quit every Claude window before updating.
Limits and open questions
The project calls itself unofficial and says it is for legitimate use only. It explicitly does not support running several accounts to stretch usage quotas, and says issues and pull requests asking for that will be closed. The bundled plugin refuses to hand work to another account because one account hit a usage limit.
This suggests the tool depends on behaviour Anthropic has not documented for this purpose: the --user-data-dir flag, and whether several concurrent sign-ins are acceptable under the terms of service. The README does not say whether Anthropic has been asked, or what happens if a future Claude Desktop release changes how sign-in state is stored. Cross-account session messaging depends on ListAgents and SendMessage, and the source gives no compatibility guarantees for those either. Star counts and fork counts are the README's own figures on 2026-11-22 and have not been independently checked here.
What it means for building apps with AI
The interesting idea is not the extra windows, it is splitting what must stay private from what should be shared. Sign-ins stay per account; prompts, skills and configuration live in one place. Anyone wiring AI into an app makes the same call about which state belongs to a user and which belongs to the project. If you are building with Draftship, the docs on using the AI chat and choosing a model cover how model selection works there. Draftship is not affiliated with this project and does not support multiple Claude accounts.
Primary sources
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.