BrowserOS neo, developed by BrowserOS, is a second browser built specifically for AI agents. It closes the gap for AI that reasons well in chat but cannot actually operate a browser: agents such as Claude Code, Cursor, and Codex take control over MCP (Model Context Protocol) while carrying over your real logged-in accounts. Because it runs separately from your everyday browser, you can delegate web tasks to agents in parallel without interrupting your own work. Everything is processed on your own machine, and the project is published as open source (AGPL-3.0). Released in July 2026, available for Mac/Windows.
Key Features
- Use your logged-in accounts as they are: Instead of a throwaway sandbox, agents work with your actual signed-in state for Gmail, calendars, work tools, and more. Importing login data from Chrome takes one click
- One-click MCP connection: Connects to major AI coding tools such as Claude Code, Cursor, Codex, and VS Code over MCP. No configuration files to write — installed agents are detected automatically
- Compressed page information to save tokens: Rather than passing raw HTML, it hands agents a compressed page representation, holding down LLM token usage while enabling browser control
- Parallel execution of multiple agents: Multiple agents run simultaneously in isolated tabs, so research, form filling, and routine tasks can proceed in parallel
- Session replay for verification: Every action is recorded locally as a scrubbable video, so you can rewind and verify or debug exactly what an agent did
- Local-first design: Data, screenshots, and session history all stay on your machine and are never sent to outside services
Pricing
| Plan | Price | Highlights |
|---|---|---|
| Free | $0 | All features available. Published as open source (AGPL-3.0) |
Pricing is as of August 2026. Check the official site for the latest information.
Pros & Cons
✅ Pros
- Completely free and open source, with the code available for inspection on GitHub
- Using logged-in accounts enables “real work on real accounts” that sandbox-style tools cannot do
- Local-only processing means no data leaves your machine, making it easier to adopt even with work accounts
- Session replay provides peace of mind by letting you verify agent actions after the fact
- One-click connection for MCP-compatible tools keeps the setup barrier low
⚠️ Cons
- Since agents get access to real accounts, managing the risk of mistakes (accidental sends, deletions, etc.) is the user’s responsibility
- Supported on Mac/Windows only, with no mobile option
- It assumes an MCP-capable agent environment (such as Claude Code), which raises the barrier for non-developers
- As a new tool released in July 2026, its features and stability are still maturing
Comparison with Similar Services
| Aspect | BrowserOS neo | Playwright MCP | Browserbase | Comet (Perplexity) |
|---|---|---|---|---|
| Form | Local second browser | Browser automation MCP server | Cloud headless browser platform | AI-integrated browser (for everyday use) |
| Runs on | Your own PC | Your own PC | Cloud | Your own PC |
| Logged-in accounts | Carried over | Possible via profiles (manual setup) | Session management built yourself | Integrated into the browser itself |
| Parallel execution | Parallel per tab | Depends on configuration | Scales in the cloud | Not intended |
| Pricing | Free (OSS) | Free (OSS) | Free tier / paid plans | Free (paid plans for advanced features) |
Who Is It For
- Developers using Claude Code or Cursor who want to delegate “real browser work” to their agents
- People who cannot send work data to external clouds and are looking for local-only browser automation
- Anyone who wants to hand routine tasks — SaaS admin panels, form filling, web research — to AI while continuing their own work in parallel
- Users who value a mechanism for recording and verifying agent actions
- Those who prefer open-source tools whose internals can be inspected
Summary
BrowserOS neo makes browser operation by AI agents a practical option through its combination of “real logged-in accounts,” “local-only processing,” and “MCP connectivity.” Unlike cloud-based browser automation, data never leaves your machine, and session replay allows after-the-fact verification, making it a good fit for work involving real accounts. Since it is free and open source, anyone with an MCP-capable environment such as Claude Code should start by trying it on a small routine task.