Agent-driven browsing over MCP
BrowserOS neo is meant to be driven by agents that speak MCP, including Claude, Cowork, Codex, and Cursor according to the homepage. You provide the task in plain language and the agent works in the browser.
BrowserOS neo is a local, open-source browser that AI agents can drive using your logged-in accounts. It is designed for interactive tasks where you want to watch the session, use your existing logins, and replay what happened if something goes wrong.

BrowserOS neo is a local, open-source browser that AI agents can control while signed into your own accounts. The site describes it as a separate browser for agent tasks, with support for MCP-driven agents and one-click import of Chrome logins.
It is designed for interactive web work where you want to watch the steps as they happen and replay the session afterward. The product is positioned for errands inside accounts like email, docs, or other logged-in sites, not for CI pipelines or test automation.
BrowserOS neo is meant to be driven by agents that speak MCP, including Claude, Cowork, Codex, and Cursor according to the homepage. You provide the task in plain language and the agent works in the browser.
The site says you can import Chrome logins in one click and work in a browser signed in as you. That lets agents operate inside accounts you already use, rather than starting from logged-out sessions.
Sessions are shown live and can be replayed afterward as a scrubbable video. This supports review, troubleshooting, and understanding what the agent actually did.
The homepage says you can run agents in parallel, with several agents sharing one visible browser. That is useful when you want multiple errands handled at the same time.
BrowserOS neo is presented as a local, open-source browser, with AGPL-3.0 noted on the comparison page. It is installed as an application rather than embedded as a framework.
The comparison page says neo is Chromium only. That narrows its scope, but also makes its position clear: it is a browser for agent work, not a multi-engine testing stack.
Use neo when an agent needs to complete work inside accounts you already control, such as email, file storage, GitHub, or other authenticated web apps.
It is suited to tasks where you want to watch the browser while the agent acts, rather than handing the job to an unattended background process.
Choose neo when you want a replayable record of what happened, so you can inspect the path the agent took if the task fails or behaves unexpectedly.
It can be used when you want more than one agent working at the same time in a shared visible browser environment.
The comparison pages position neo as a better fit than Playwright for errands that are not meant to become committed tests, especially when working in real accounts matters more than test infrastructure.
No. The comparison page says it is a browser application and explicitly notes that it is not a test runner and is not built for CI or assertions.
Yes. The site says you can import Chrome logins in one click and use the browser signed in as you.
The homepage names Claude, Cowork, Codex, and Cursor, and the comparison pages describe it as usable by agents that speak MCP.
Yes. The site says sessions can be watched live and replayed afterward as a scrubbable video.
The pricing URL returns a page-not-found response, so no pricing plan details or limits are confirmed on the source pages provided.