What is the Claude Code integration?
Claude Code can use a browser workflow through a skill and its available tools, or through a configured MCP server. Remote Browser provides the browser runtime for that workflow. A skill describes how to perform the task; the installed tools make the actual connection and browser actions possible.
The Claude Code skills documentation describes instruction packages built around SKILL.md. Remote Browser's own skill setup supplies the product-specific installation and endpoint instructions.
Why give Claude Code a remote browser?
A code change often needs verification in a running application. A remote browser can provide a clearly identified test account and a human-visible session outside the coding worker. This is useful for checking a deployed UI, investigating an authenticated navigation problem, or preserving evidence for a code review.
Keep the task precise. Specify the application, account boundary, expected result, and whether the agent may submit changes. A browser connection does not establish that every action inside an authenticated account is part of the task.
Quick start
Follow the Remote Browser Agent skill instructions. The documented installer supports a Claude skills location when the expected Claude home directory exists; verify the destination in your setup rather than assuming installation into another agent's directory is sufficient.
Provide REMOTE_BROWSER_CDP_URL securely in the environment used by the browser tool process. Then try a small read-only task:
Use the Remote Browser skill and the configured browser endpoint.
Open https://example.com and report the page title and final URL.
Keep the endpoint secret. Use the intended session and leave unrelated tabs alone.Verify the browser result in the viewer where available. If the skill is not discovered, check its installation path and the client's skill loading behavior before debugging the browser connection.
Connection methods
The skill route packages the Remote Browser workflow and connection conventions. An MCP route instead exposes tools from a separately configured server, such as Playwright MCP. Choose the interface your team can maintain and avoid configuring two uncontrolled writers against one browser.
Neither route makes the browser share Claude Code's working directory. Test file upload and download transfer when reproducing an issue involving local fixtures or generated artifacts.
Use cases
- Validate a deployed UI against a concrete acceptance criterion.
- Reproduce a bug with an authorized test account.
- Inspect a dashboard while a human assists with authentication renewal.
- Collect screenshots and an account-safe report for a code review.
For repeatable evaluation, keep the application version and test data fixed. Record whether the result was verified on the page, not merely described by the agent.
FAQ
Does Claude Code need to run on the browser machine?
The remote connection separates the worker from the browser runtime. Network access and credentials still need to be available to the tool that attaches.
Can a profile bypass two-factor authentication?
A profile can retain supported state. The website may still require verification; use an authorized human takeover and resume only after the account is confirmed.
What should happen when the task ends?
Report the observed result and relevant evidence, then follow the agreed session release policy. Provider-side cleanup remains a separate operational responsibility.