CLI Quickstart

Install the BAP CLI and run your first browser commands

The BAP CLI lets AI coding agents automate browsers from the shell. It defaults to installed Chrome, keeps a persistent daemon running, and supports composite actions that batch multiple steps into one command.

Installation

1. Install the CLI globally

   npm i -g @browseragentprotocol/cli
   

   Or run without installing:

   npx @browseragentprotocol/cli goto https://example.com
   
2. Install Playwright browsers (if needed)

   BAP prefers your installed Chrome. If Chrome is not available: bash npx playwright install chromium

3. Install the skill file for your AI agent

   bash bap install-skill This auto-detects 13 AI agent platforms (Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot, Windsurf, Roo Code, and more) and installs the SKILL.md guidance file.

4. Run the guided demo

   bap demo
   

   Walks you through navigate, observe, click, and screenshot step by step.

First Commands

# Open a page and observe interactive elements
bap goto https://example.com --observe

# Click a link by visible text
bap click text:"More information..."

# Take a screenshot
bap screenshot

The BAP server starts automatically on first use and stays running as a background daemon. Use bap close-all to stop it.

Composite Actions

The killer feature of BAP CLI is bap act — batch multiple browser steps in one command:

Single command (BAP)

bap act fill:label:"Email"="[e-mail]" \
        fill:label:"Password"="secret123" \
        click:role:button:"Sign in"

Step syntax: action:selector=value for fill/type, action:selector for click/check/hover.

Observation

# Interactive elements (default max 50)
bap observe

# Full accessibility tree
bap observe --full

# Form fields only
bap observe --forms

# Incremental: only changes since last observation
bap observe --diff

# Minimal response (fewer tokens)
bap observe --tier=minimal

Fused Operations

Combine navigation or actions with observation in a single server call:

# Navigate + observe in one roundtrip
bap goto https://example.com --observe

# Act + observe in one roundtrip
bap act click:text:"Next" --observe

# Act + observe with minimal response tier
bap act click:e3 --observe --tier=interactive

Structured Extraction

# Extract specific fields
bap extract --fields="title,price,rating"

# Extract a list of items
bap extract --list="product"

# Extract using a JSON schema
bap extract --schema=product.json

Sessions and Tabs

# Named sessions (persist across commands)
bap -s=research goto https://example.com
bap -s=shopping goto https://store.com

# List sessions and tabs
bap sessions
bap tabs

# Tab management
bap tab-new https://other.com
bap tab-select 2

Global Options

-s=<name>              Named session
-p, --port <N>         Server port (default: 9222)
-b, --browser <name>   chrome, chromium, firefox, webkit, edge
--headless             Headless mode for CI/background runs
--no-headless          Show browser window (default)
--profile <path>       Chrome profile directory
--no-profile           Fresh browser, no user profile
--stealth              Reduce bot detection fingerprint
--connect              Auto-attach to running Chrome via CDP
-v, --verbose          Verbose output

Output

Commands produce concise, agent-friendly stdout:

### Page
- URL: https://example.com/dashboard
- Title: Dashboard
### Snapshot
[Snapshot](.bap/snapshot-2026-02-16T19-30-42.yml)

Files are saved to .bap/ in the current directory: snapshots (.yml), screenshots (.png), extractions (.json).

Migrating from playwright-cli

BAP is a drop-in replacement. All e<N> refs from snapshots work identically:

playwright-cli	bap
playwright-cli open [url]	bap open [url]
playwright-cli click e15	bap click e15
playwright-cli fill e5 "text"	bap fill e5 "text"
playwright-cli snapshot	bap snapshot
playwright-cli screenshot	bap screenshot

BAP adds composite actions, semantic selectors, smart observation, and structured extraction on top.