Let agents test your code in a real browser.
One command scans your unstaged changes or branch diff, generates a test plan, and runs it against a live browser.
npx -y expect-cli@latest initUsage: expect-cli [options] [command]
Options:
-m, --message <instruction> natural language instruction for what to test
-f, --flow <slug> reuse a saved flow by slug
-y, --yes skip plan review, run immediately
-a, --agent <provider> agent provider to use (claude, codex, copilot, gemini, cursor, opencode, or droid)
-t, --target <target> what to test: unstaged, branch, or changes (default: changes)
--verbose enable verbose logging
-v, --version print version
-h, --help display help
Commands:
init install expect globally and set up skill
Examples:
$ expect-cli open interactive TUI
$ expect-cli -m "test the login flow" -y plan and run immediately
$ expect-cli --target branch test all branch changes
$ expect-cli --target unstaged test unstaged changes
Set `NO_TELEMETRY=1` to disable analytics events.
flowchart LR
A["Scan changes\n(git diff)"] --> B["Generate plan\n(AI agent)"]
B --> C["Run in browser\n(Playwright)"]
C --> D["Report\n(pass/fail)"]
Expect reads your unstaged changes or branch diff, sends them to an AI agent, and generates a step-by-step test plan describing how to validate the changes. You review and approve the plan in an interactive TUI, then the agent executes each step against a live browser - using your real login sessions so there's no manual auth setup. Every session is recorded so you can replay exactly what happened.
Pass -y to skip plan review and run headlessly in CI. Exits 0 on success, 1 on failure.
Most of these tools let an agent click around a browser. Expect does that too, but it also reads your diff, writes a test plan, runs it with your real cookies across multiple browsers, and gives you a live view + replay of every session.
| Tool | What it does | Auth | Browsers | Recording | Diff-aware | Live view | CI |
|---|---|---|---|---|---|---|---|
| Expect | Reads your diff, generates a test plan, runs it in a real browser | Extracts cookies from Chrome/Firefox/Safari profiles | Multi-browser | rrweb replay viewer | Yes | Yes | Yes |
| Playwright CLI | Traditional test runner, you write and maintain tests by hand | Manual storageState JSON |
Chromium, Firefox, WebKit | Trace Viewer | --only-changed filters existing tests, doesn't generate new ones |
No | Yes |
| Playwright MCP | Exposes Playwright as MCP tools, any LLM can drive a browser via accessibility snapshots | None, starts logged out | Chromium only | No | No | No | Yes |
| Claude in Chrome | Extension connects Claude Code to your running Chrome | Inherits your browser session | Whichever Chrome window you connect | GIF only | No | No | No |
| Playwriter | Extension connects any MCP client to your running Chrome | Inherits your browser session | Chrome only (needs extension) | Video capture | No | No | No |
| Agent Browser | Fast Rust CLI for headless browser automation, supports cloud providers | None locally, Kernel cloud profiles | Chromium only locally | No | No | No | Yes |
| Chrome DevTools MCP | Google's MCP server for DevTools: perf tracing, network inspection, Lighthouse | None, isolated profile | Chrome only | Experimental screencast | No | No | Yes |
| Dev Browser | Runs sandboxed JS against Playwright browsers | Can connect to running Chrome, no extraction for headless | Chromium only | No | No | No | Yes |
Playwright codegen |
Records your manual clicks and outputs test code you maintain | None | One at a time | No | No | No | No |
Expect works with the following coding agents via the Agent Client Protocol (ACP):
| Agent | Flag | Install |
|---|---|---|
| Claude Code | -a claude |
npm install -g @anthropic-ai/claude-code |
| Codex | -a codex |
npm install -g @openai/codex |
| GitHub Copilot | -a copilot |
npm install -g @github/copilot |
| Gemini CLI | -a gemini |
npm install -g @google/gemini-cli |
| Cursor | -a cursor |
cursor.com |
| OpenCode | -a opencode |
npm install -g opencode-ai |
| Factory Droid | -a droid |
npm install -g droid |
Expect auto-detects which agents are installed on your PATH. If multiple are available, it defaults to the first one found. Use -a <provider> to pick a specific agent.
Want to try it out? Check out our demo.
Find a bug? Head over to our issue tracker and we'll do our best to help. We love pull requests, too!
We expect all contributors to abide by the terms of our Code of Conduct.
→ Start contributing on GitHub
FSL-1.1-MIT © Million Software, Inc.