A small thing I learned about Claude in Chrome and how to make it more useful.
One of the limitations I found with the original Claude Code in Web was the lack of commands and agents. Granted, you could add some to the local .claude of the project, but that meant either losing access to your personal configuration, or having to duplicate that across projects, along with making other team members agree on what was useful to keep in there.
When I tested Claude in Chrome, I had the same feeling. It was a short test and I hadn’t read the documentation, but the result felt useful while lacking some punch.
Then I discovered the --chrome flag for claude, which made the browser workflow much more useful.
First, ensure you have the following available:
- Claude Pro/Max subscription
- Claude in Chrome extension
- Claude Code CLI
Then try the following:
- Create a repository with a
.claudefolder and all relevant skills, agents, and commands for your use case - Within that repository, run
claude --chrome
This command will use an existing Chrome window and profile, and the Claude window it opens will have access to the commands and skills in the repository.
This is useful for two reasons:
First, the fact that it reuses an existing window, including the profile, means you can use it with profiles managed remotely that have access to restricted resources. This is handy for automation of tasks, or even just to avoid having to log into a service each time, as the session persists.
Second, it means that you can protect your context window by loading only those skills and tools relevant to the tasks you have in mind. Managing context is an important skill, and this helps.
As an extra tip, you can use other flags to, for example, tailor the system prompt of Claude:
claude --chrome --append-system-prompt "$(cat "specific-instructions.md")"This gives the browser session a focused set of tools and behaviour for the task, without adding the same configuration to every project.