Skip to main content
Reference for the claude-science command: its subcommands, the serve flags, the single-use login link, and the environment variables you can set.
claude-science serve starts Claude Science and opens the web app in your browser at a single-use login link. On Linux, everyday use is that one command. The others manage the running program: they mint login links, report status, follow logs, install updates, and merge data directories. On Windows, the installer adds the command to your PATH for new terminals. There the app window you open from the Start menu is the everyday way in, and the commands below manage the same running program.

​
Commands

import has no preview and no undo. Back up the data directory before you run it; running the same import a second time is safe.

​
Roll back to an earlier build

claude-science update --to <build> installs one specific build. <build> is an 8-character build ID such as 3f9a01bc, not a version number. claude-science update --check prints the ID of the installed build as Current. Claude Science checks for updates in the background and shows Update available when one is ready. Note the Current ID before you choose Restart to update, so you know which build to go back to. If a newer build has already changed how your data directory stores sessions, the command refuses an older build and installs nothing, because an older build can permanently lose part of every session it writes to. The refusal names a flag that overrides it, --accept-layout-downgrade, and says when that is safe. If you cannot tell from the message whether it is safe for you, do not add the flag.
An older build that does install can still refuse to start if a newer build has already updated the database in your data directory. When it does, it stops at startup and changes nothing in your data. At a terminal, it says the data was written by a NEWER version. To start the app again, run claude-science update, which installs the latest build.

​
Global flags

These two work on every command. On Windows, ~ in the defaults below is your user profile folder, %USERPROFILE%. When serve starts, it prints a line of the form Web UI → http://localhost:<port>/?nonce=.... The nonce is a one-time password: it signs one browser tab in and then expires, about three minutes after it is printed. The signed-in tab stays signed in until you restart the program.
You never need to keep a link. claude-science open mints a fresh one and opens it in your browser whenever you want to sign in again. The app listens on 127.0.0.1 unless you change --host, so it is reachable only from the machine it runs on. For a machine you reach over SSH, first forward two ports from your computer, the web app port and the preview port (by default the web app port plus one), as Run on a remote Linux server shows. Then run claude-science url on that machine and open the link it prints in your own browser.

​
Flags for serve

​
Dangerous flags

claude-science serve accepts two flags that each turn off a protection.
  • --dangerously-no-sandbox turns the sandbox off. On macOS and Linux, the code Claude runs then has full read and write access to your home directory and unrestricted network access. On Windows, code cells do not run at all while the flag is set.
  • --dangerously-skip-approvals approves permission cards for you without showing them, until you restart without the flag. That includes Claude’s requests to run code, reach network hosts, open folders, and use connector tools. Cards still show for a connector tool you set to Ask each time and for requests only you can answer, such as entering an SSH password, permanently deleting artifacts, or spending usage credits. Questions Claude asks you still appear.
These flags are acceptable only on a machine that is already isolated, such as a container or a disposable virtual machine. Never use them with data or prompts that came from someone else, because Claude may follow instructions hidden in them. Neither belongs in everyday use.
If your organization manages the network allowlist, the app ignores --dangerously-no-sandbox and keeps the sandbox on. If the app learns of that setting only after it has started without the sandbox, it pauses new sessions and messages until you restart it.

​
Environment variables

DO_NOT_TRACK, set to any value other than 0 or false, turns usage analytics and error reports off. It is the same switch as disable_telemetry = true in the configuration file. GITHUB_TOKEN (or GH_TOKEN) is optional and is used only against api.github.com, to lift the rate limit when you install a skill from a GitHub repository. Claude Science also reads the standard proxy variables (HTTPS_PROXY, HTTP_PROXY, NO_PROXY, and ALL_PROXY); see Use Claude Science on a corporate network. The proxy address variables are the one case where the environment overrides the configuration file, and NO_PROXY is merged with the no_proxy key rather than replacing it. Every other setting belongs in the configuration file. An app started from the macOS Dock or Finder, or from the Windows Start menu, does not see variables exported in a terminal. The macOS app reads the env file in the data directory (by default ~/.claude-science/env) when it starts, so put the variable there as a KEY=VALUE line, then quit and reopen the app. On Windows, set it as a user environment variable, then quit and reopen the app. See How the environment variables reach the app.

​
See also

Run on a remote Linux server

Install Claude Science on a server and use it from your own browser through an SSH tunnel.

Configuration file reference

Find the config.toml file and set its network-related keys.

Remote compute clusters

Connect an SSH host and run jobs on it.