webcli

Headless browser CLI for AI coding agents. Control a real Chromium browser from the command line.

Built for Claude Code and similar AI-powered development tools that need to interact with web pages programmatically.

Links

npm	npmjs.com/package/@erdinccurebal/webcli
Website	webcli.erdinc.curebal.dev
GitHub	github.com/erdinccurebal/webcli
ClawHub	clawhub.ai/skills/webcli
Skills.sh	npx skills add erdinccurebal/webcli

Features

• 55+ commands — navigate, read, interact, snapshot, screenshot, PDF, console, storage, state management
• Accessibility snapshot — get the page's accessibility tree with deterministic refs (@e1, @e2...) for reliable element targeting
• Persistent daemon — browser stays open between commands, no startup overhead
• Multi-tab — manage multiple named browser tabs simultaneously
• Anti-detection — realistic user-agent, webdriver flag removal, configurable viewport
• Session persistence — save/load full browser state (cookies, localStorage, sessionStorage)
• Device emulation — emulate mobile devices (iPhone, iPad, Pixel, etc.)
• JSON output — structured output for programmatic consumption
• Headed/headless — toggle via config for sites with bot detection
• Agent skill — works as a skill for Claude Code, Cursor, Copilot, Codex, Windsurf and 30+ other agents

Installation

As npm package

npm install -g @erdinccurebal/webcli
npx playwright install chromium

As agent skill

Skills.sh (Claude Code, Cursor, Copilot, Codex, Windsurf + 30 more):

npx skills add erdinccurebal/webcli

ClawHub (OpenClaw):

clawhub install webcli

Quick Start

webcli go https://example.com # Navigate (auto-starts daemon)
webcli source# Read page text
webcli links # List all links
webcli click "More information..."# Click by visible text
webcli snapshot # Get accessibility tree with refs
webcli click @e1 # Click using snapshot ref
webcli screenshot -o page.png # Take screenshot
webcli stop # Stop daemon

Commands

Navigation

Command	Description
webcli go <url>	Navigate to URL
webcli back	Go back in history
webcli forward	Go forward in history
webcli reload	Reload current page

Options for go: -w, --wait <strategy> — domcontentloaded (default), networkidle, load

Reading

Command	Description
webcli source	Get visible text content of the page
webcli html <selector>	Get innerHTML of a specific element
webcli attr <selector> <attribute>	Get an element's attribute value
webcli links	List all links (text + href)
webcli forms	List all forms with their inputs
webcli eval <js>	Execute JavaScript and return result
webcli title	Get the page title
webcli url	Get the current URL
webcli value <selector>	Get the value of an input/textarea
webcli count <selector>	Count elements matching a selector
webcli box <selector>	Get element bounding box (x, y, width, height)
webcli styles <selector> [props...]	Get computed CSS styles
webcli visible <selector>	Check if element is visible (bool)
webcli enabled <selector>	Check if element is enabled (bool)
webcli checked <selector>	Check if checkbox is checked (bool)

Interaction

Command	Description
webcli click <text>	Click element by visible text
webcli clicksel <selector>	Click element by CSS selector
webcli dblclick <text>	Double-click element by visible text
webcli hover <selector>	Hover over an element
webcli focus <selector>	Focus element by CSS selector
webcli type <text>	Type text with keyboard
webcli fill <selector> <value>	Fill an input field
webcli select <selector> <value>	Select a dropdown option
webcli press <key>	Press a keyboard key (Enter, Tab, Escape...)
webcli check <selector>	Check a checkbox
webcli uncheck <selector>	Uncheck a checkbox
webcli drag <from> <to>	Drag and drop between two selectors
webcli upload <selector> <file>	Upload a file to an input element
webcli scroll <direction> [amount]	Scroll the page (up/down/left/right)

All interaction commands support snapshot refs: webcli click @e1, webcli fill @e3 "value"

Accessibility Snapshot

Command	Description
webcli snapshot	Get accessibility tree with interactive element refs
webcli snapshot --all	Include non-interactive elements
webcli snapshot --max-depth <n>	Limit tree depth

The snapshot assigns deterministic refs (@e1, @e2, ...) to interactive elements. Use these refs in other commands for reliable element targeting:

webcli snapshot # Get tree with refs
webcli click @e1 # Click using ref
webcli fill @e3 "hello"# Fill input using ref
webcli hover @e5 # Hover using ref

Waiting

Command	Description
webcli wait <selector>	Wait for CSS selector to become visible
webcli waitfor <text>	Wait for text to appear on page
webcli sleep <ms>	Sleep for specified milliseconds

Screenshots & PDF

Command	Description
webcli screenshot	Take a full-page screenshot
webcli screenshot --no-full	Viewport-only screenshot
webcli screenshot --annotate	Screenshot with numbered interactive elements
webcli pdf	Save page as PDF

Options: -o, --output <file> for custom output path.

Console & Errors

Command	Description
webcli console on	Start capturing console messages
webcli console off	Stop capturing
webcli console	Show captured console messages
webcli errors	Show page errors

Dialog Handling

Command	Description
webcli dialog accept [text]	Accept a browser dialog (alert/confirm/prompt)
webcli dialog dismiss	Dismiss a browser dialog

Frame (iframe) Switching

Command	Description
webcli frame <selector>	Switch to an iframe
webcli frame main	Switch back to main frame

Storage (localStorage)

Command	Description
webcli storage get [key]	Get localStorage value(s)
webcli storage set <key> <value>	Set a localStorage value
webcli storage clear	Clear all localStorage

State Management

Command	Description
webcli state save [file]	Save full browser state (cookies + storage) to JSON
webcli state load <file>	Load browser state from JSON

Cookies & Browser

Command	Description
webcli cookie export	Export cookies as JSON
webcli cookie import <file>	Import cookies from JSON file
webcli viewport <width> <height>	Change viewport size
webcli useragent <string>	Change user agent
webcli device <name>	Emulate a device (e.g. "iPhone 14", "iPad Pro 11")
webcli network [on|off]	Toggle request/response logging

Tab & Daemon Management

Command	Description
webcli tabs	List all open tabs
webcli quit	Close a tab
webcli status	Show daemon status (PID, uptime, tabs)
webcli stop	Stop the daemon and close browser

Global Options

Option	Description
-t, --tab <name>	Target tab (default: "default")
--json	Output as JSON
--timeout <ms>	Command timeout (default: 30000)
--verbose	Enable debug logging

Configuration

Create ~/.webcli/config.json:

{
"headless": true,
"browser": "chromium",
"userDataDir": "~/.webcli/browser-data",
"viewport": { "width": 1280, "height": 800 },
"locale": "en-US",
"timezoneId": "America/New_York",
"geolocation": { "latitude": 40.7128, "longitude": -74.0060 },
"colorScheme": "dark",
"offline": false,
"extraHeaders": { "X-Custom": "value" },
"waitAfterClick": 500,
"waitAfterPress": 300,
"idleTimeout": 300000,
"defaultTimeout": 30000
}

Set "headless": false to use headed mode (required for sites with aggressive bot detection like X/Twitter).

Architecture

CLI Client (webcli)
|
|-- Unix Socket (~/.webcli/daemon.sock)
|
v
Daemon (background process)
|
|-- Playwright Browser (Chromium/Firefox/WebKit)
| |-- Tab: "default" -> Page
| |-- Tab: "search" -> Page
| +-- Tab: "login" -> Page
|
|-- Tab Manager (named page instances)
|-- Handler Registry (action dispatch)
|-- Ref Store (snapshot @e1 refs -> ElementHandle)
+-- Auto-shutdown after 5min idle

The daemon starts automatically on the first command and persists in the background. All commands communicate via a Unix domain socket using newline-delimited JSON.

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT