Dev Browser

Dev Browser gives AI coding agents like Claude Code, Codex, and Amp direct control over a real browser — without the usual MCP overhead or fragile plugin setups. It runs scripts inside a QuickJS WASM sandbox, so there's zero risk of host access or accidental damage to your system. The killer feature is persistent named pages. You call `browser.getPage('main')`, navigate to a URL, and that page stays alive across multiple script executions. No re-launching, no lost state. Your agent can navigate once, then interact across dozens of follow-up commands. Under the hood, you get the full Playwright Page API — `goto`, `click`, `fill`, `locator`, `evaluate`, `screenshot`, and the AI-specific `page.snapshotForAI()` method that returns structured page data optimized for LLM consumption. Performance tells the real story. In benchmarks against Playwright MCP, Playwright Skill, and the Claude Chrome Extension, Dev Browser completed tasks in 3 minutes 53 seconds at $0.88 per task — roughly 3x faster and 3x cheaper than the Chrome Extension alternative (12m54s / $2.81). It also used fewer turns (29 vs 80), which means less back-and-forth and tighter agent loops. Setup takes two commands: `npm install -g dev-browser && dev-browser install`. After that, just tell your agent to run `dev-browser --help` — the help output includes a full LLM usage guide with examples and API reference. No config files, no environment variables, no plugin marketplace hoops. It auto-connects to your running Chrome instance (via `chrome://inspect/#remote-debugging`) or launches a fresh Chromium. Headless mode works out of the box for CI/CD pipelines and automated testing. File I/O is sandboxed to `~/.dev-browser/tmp/` for screenshots and data exports. Created by Sawyer Hood, who also built Do Browser. The project has 3,949 GitHub stars and is MIT licensed. Windows is not currently supported.