vercel-labs /agent-browser

Browser automation CLI for AI agents. Fast native Rust CLI.

(opens in a new tab)

Installation

Global Installation (recommended)

Installs the native Rust binary:

npm install -g agent-browser
agent-browser install  # Download Chrome from Chrome for Testing (first time only)

Project Installation (local dependency)

For projects that want to pin the version in package.json:

npm install agent-browser
agent-browser install

Then use via package.json scripts or by invoking agent-browser directly.

Homebrew (macOS)

brew install agent-browser
agent-browser install  # Download Chrome from Chrome for Testing (first time only)

Cargo (Rust)

cargo install agent-browser
agent-browser install  # Download Chrome from Chrome for Testing (first time only)

From Source

Requires Node.js 24+, pnpm 11+, and Rust.

git clone https://github.com/vercel-labs/agent-browser
cd agent-browser
pnpm install
pnpm build
pnpm build:native   # Requires Rust (https://rustup.rs)
pnpm link --global  # Makes agent-browser available globally
agent-browser install

Linux Dependencies

On Linux, install system dependencies:

agent-browser install --with-deps

This exits nonzero if the package manager cannot install every required browser library.

Updating

Upgrade to the latest version:

agent-browser upgrade

Detects your installation method (npm, Homebrew, or Cargo) and runs the appropriate update command automatically.

Requirements

• Chrome - Run agent-browser install to download Chrome from Chrome for Testing (opens in a new tab) (Google’s official automation channel). Existing Chrome, Brave, Playwright, and Puppeteer installations are detected automatically. No Playwright or Node.js required for the daemon.
• Node.js 24+ and pnpm 11+ - Only needed when building from source.
• Rust - Only needed when building from source (see From Source above).

Quick Start

agent-browser open example.com
agent-browser snapshot                    # Get accessibility tree with refs
agent-browser click @e2                   # Click by ref from snapshot
agent-browser fill @e3 "test@example.com" # Fill by ref
agent-browser get text @e1                # Get text by ref
agent-browser screenshot page.png
agent-browser close

Clicks fail early when another element covers the target’s click point, for example a consent banner or modal. Dismiss or interact with the reported covering element, then take a fresh snapshot before retrying the original ref.

Headless Chromium screenshots hide native scrollbars for consistent image output. Pass --hide-scrollbars false when launching to keep native scrollbars visible.

Traditional Selectors (also supported)

agent-browser click "#submit"
agent-browser fill "#email" "test@example.com"
agent-browser find role button click --name "Submit"

Commands

Core Commands

agent-browser open                    # Launch browser (no navigation); stays on about:blank

Get Info

agent-browser get text <sel>          # Get text content
agent-browser get html <sel>          # Get innerHTML
agent-browser get value <sel>         # Get input value
agent-browser get attr <sel> <

Read Agent-Friendly Text

agent-browser read
agent-browser read https://example.com/article
agent-browser read https://example.com/article --filter overview
agent-browser read https://example.com/article --outline
agent-browser read https://docs.example.com --llms index --filter auth
agent-browser read https://docs.example.com --llms full --filter auth
agent-browser read

read fetches a URL without launching Chrome. Omit the URL to read the rendered DOM of the active tab in the current browser session, including browser auth state and client-side updates. Explicit URL reads send Accept: text/markdown by default, try the same URL with .md appended when the first response is not markdown, walk ancestor paths toward / to find the nearest llms.txt for a matching docs link, print markdown or plain text when available, and fall back to readable text extracted from HTML. --llms and --require-md with no URL use the active tab URL because they depend on HTTP resources. read does not read llms-full.txt unless you ask for it.

Options: --raw prints the response body without HTML extraction, --require-md fails unless the server returns Content-Type: text/markdown, --outline prints a compact heading outline for one page, --llms index prints a compact nearest-ancestor llms.txt link list, --llms full reads the nearest-ancestor llms-full.txt, --filter <text> narrows page sections, llms links/sections, or outline headings, and --timeout <ms> changes the request timeout. Global safeguards such as --allowed-domains, --content-boundaries, and --max-output also apply to read fetches and output.

Check State

agent-browser is visible <sel>        # Check if visible
agent-browser is enabled <sel>        # Check if enabled
agent-browser is checked <sel>        # Check if checked

Find Elements (Semantic Locators)

agent-browser find role <role> <action> [value]       # By ARIA role
agent-browser find text <text> <action> [value]       # By text content
agent-browser find label <labe

Actions: click, fill, check, hover, text

Options: --name <name> (filter role by accessible name), --exact (exact, case-sensitive match; for role it applies to the accessible name, whose default is a case-insensitive substring)

Examples:

agent-browser find role button click --name "Submit"
agent-browser find role heading text --name "Skills"     # implicit roles work: <h2>=heading, <ul>=list, top-level <header>=banner
agent-browser find text "Sign In" click
agent-browser find label "Email" fill "test@test.com"
agent-browser find first ".item" click
agent-browser

Wait

agent-browser wait <selector>         # Wait for element to be visible
agent-browser wait <ms>               # Wait for time (milliseconds)
agent-browser wait --text "Welcome"   # Wait for text to appear (substring match)
agent-browser wait --url "**/dash"    # Wait for URL pattern
agent-browser wait --load networkidle # Wait for load state

Load states: load, domcontentloaded, networkidle

Batch Execution

Execute multiple commands in a single invocation. Commands can be passed as quoted arguments or piped as JSON via stdin. This avoids per-command process startup overhead when running multi-step workflows.

# Argument mode: each quoted argument is a full command
agent-browser batch "open https://example.com" "snapshot -i" "screenshot"

# With --bail to stop on first error
agent-browser batch --bail "open https://example.com" "click @e1" "screenshot"

# Stdin mode: pipe commands as JSON
echo '[
  ["open", "https://example.com"],
  ["snapshot", "-i"],
  ["click", "@e1"],
  ["screenshot", "result.png"]
]' | agent-browser batch --json

Clipboard

agent-browser clipboard read                      # Read text from clipboard
agent-browser clipboard write "Hello, World!"     # Write text to clipboard
agent-browser clipboard copy                      # Copy current selection (Ctrl+C)
agent-browser clipboard paste                     # Paste from clipboard (Ctrl+V)

Mouse Control

agent-browser mouse move <x> <y>      # Move mouse
agent-browser mouse down [button]     # Press button (left/right/middle)
agent-browser mouse up [button]       # Release button
agent-browser mouse wheel <dy> [dx]   # Scroll wheel

Browser Settings

agent-browser set viewport <w> <h> [scale]  # Set viewport size (scale for retina, e.g. 2)
agent-browser set device <name>       # Emulate device ("iPhone 14")
agent-browser set geo <lat> <lng>     # Set geolocation
agent-browser set

Cookies & Storage

agent-browser cookies                 # Get all cookies
agent-browser cookies set <name> <val> # Set cookie
agent-browser cookies set --curl <file> # Import cookies from a Copy-as-cURL dump,
                                        # JSON array, or bare Cookie header (auto-detected)
agent-browser cookies clear           # Clear cookies

Network

agent-browser network route <url>              # Intercept requests
agent-browser network route <url> --abort      # Block requests
agent-browser network route <url> --body

Tabs & Windows

agent-browser tab                              # List tabs (shows `tabId` and optional label)
agent-browser tab new [url]                    # New tab (optionally with URL)
agent-browser tab new --label docs [url]       # New tab with a user-assigned label
agent-browser tab <t<N>|label>                 # Switch to a tab by id or label
agent-browser tab close [t<N>|

Tab ids are stable strings of the form t1, t2, t3. They’re never reused within a session, so scripts and agents can keep referring to the same tab even after other tabs are opened or closed. Positional integers like tab 2 are not accepted; the t prefix disambiguates handles from indices and mirrors the @e1 convention used for element refs.

You can also assign a memorable label (docs, app, admin) and use it interchangeably with the id. Labels are never auto-generated and never rewritten on navigation — they’re yours to name and keep:

agent-browser tab new --label docs https://docs.example.com
agent-browser tab docs               # switch to the docs tab
agent-browser snapshot               # populate refs for docs
agent-browser click @e3              # click uses docs's refs
agent-browser tab close docs         # close by label

Switching to a tab discarded by Chrome’s Memory Saver reactivates it, since a discarded tab has no renderer to drive. Reactivation reloads the discarded page and resets its unsaved state, and the switch result reports "revived": true. A tab whose page is paused by a JavaScript dialog is alive rather than discarded, so the switch leaves it untouched and reports "dialogBlocked": true; resolve the dialog with dialog accept or dialog dismiss before interacting. Closing the active tab onto a discarded successor revives it the same way and reports "activeTabRevived": true.

Frames

agent-browser frame <sel>             # Switch to iframe
agent-browser frame main              # Back to main frame

Dialogs

agent-browser dialog accept [text]    # Accept (with optional prompt text)
agent-browser dialog dismiss          # Dismiss
agent-browser dialog status           # Check if a dialog is currently open

By default, alert and beforeunload dialogs are automatically accepted so they never block the agent. confirm and prompt dialogs still require explicit handling. Use --no-auto-dialog (or AGENT_BROWSER_NO_AUTO_DIALOG=1) to disable automatic handling.

When a JavaScript dialog is pending, all command responses include a warning field with the dialog type and message.

Diff

agent-browser diff snapshot                              # Compare current vs last snapshot
agent-browser diff snapshot --baseline before.txt        # Compare current vs saved snapshot file
agent-browser diff snapshot --selector "#main" --compact # Scoped snapshot diff
agent-browser diff screenshot --baseline before.png      # Visual pixel diff against baseline
agent-browser

Debug

agent-browser trace start             # Start recording trace
agent-browser trace stop [path]       # Stop and save trace
agent-browser profiler start          # Start Chrome DevTools profiling
agent-browser profiler stop [path]    # Stop and save profile (.json)
agent-browser console

Navigation

agent-browser back                    # Go back
agent-browser forward                 # Go forward
agent-browser reload                  # Reload page
agent-browser pushstate <url>         # SPA client-side nav; auto-detects window.next.router.push,
                                      # falls back to history.pushState + popstate

Pre-navigation setup

Some flows (SSR debug, auth cookies for protected origins, init scripts) need state set up before the first navigation. Use open with no URL to launch the browser, then stage cookies / routes / init scripts, then navigate. batch sends it all in one CLI call:

agent-browser batch \
  '["open"]' \
  '["network","route","*","--abort","--resource-type","script"]' \
  '["cookies","set","--curl","cookies.curl","--domain","localhost"]' \
  '["navigate","http://localhost:3000/target"]'

Without batch the same sequence is three commands that all reuse the same daemon (fast, but not one turn).

React / Web Vitals

Agent-browser ships with first-class React introspection and universal Web Vitals metrics. The React commands need the React DevTools hook installed at launch; Web Vitals and pushstate are framework-agnostic.

agent-browser open --enable react-devtools <url>   # Launch with React hook installed
agent-browser react tree                           # Full component tree
agent-browser react inspect <fiberId>              # props, hooks, state, source
agent-browser react renders start                  # Begin fiber render recording
agent-browser react

Each react ... subcommand requires --enable react-devtools to have been passed at launch (the React DevTools installHook.js is embedded in the binary). Without it the commands error with `React DevTools hook not installed

• relaunch with --enable react-devtools`.

Works on any React app — Next.js, Remix, Vite+React, CRA, TanStack Start, React Native Web, etc. vitals and pushstate are framework-agnostic. vitals prints a summary by default; pass --json for the full structured payload.

Accessibility audits

Run an axe-core (opens in a new tab) accessibility audit against the current page or a URL. The axe-core engine is embedded in the binary, so it works offline and under strict CSP. It runs private partial audits across the page’s frame tree and merges serialized results without page messaging, so page-provided window.axe values remain intact and iframe violations retain their frame selector paths. Accessibility audits require a CDP browser and are not available with Safari or iOS WebDriver sessions.

agent-browser a11y                                 # Audit the current page
agent-browser a11y https://example.com             # Navigate, then audit
agent-browser a11y --tags wcag2a,wcag2aa           # Only rules with these axe tags
agent-browser a11y --selector "#main"              # Scope the audit to a subtree
agent-browser a11y example.com --json              # Full structured results

The default output lists each violation with its impact, rule id, fix guidance URL, and the CSS selectors of failing nodes:

url: https://example.com/
axe-core: 4.12.1  violations: 2  incomplete: 0  passes: 24

[critical] image-alt: Images must have alternative text (3 nodes)
  https://dequeuniversity.com/rules/axe/4.12/image-alt
  - img.hero
  - #logo > img
  - footer img
[serious] color-contrast: Elements must meet minimum color contrast ratio thresholds (1 node)
  https://dequeuniversity.com/rules/axe/4.12/color-contrast
  - .nav a.muted

--json returns the same data structured for automation (counts, violations, incomplete, each violation’s nodes with target, html, and failureSummary). Each target preserves axe’s selector path arrays, including nested arrays for shadow DOM boundaries. Rules that axe could not evaluate automatically are reported under incomplete for manual review.

Init scripts

agent-browser open --init-script <path>           # Register page init script before first navigation
                                                  # (repeatable; also AGENT_BROWSER_INIT_SCRIPTS env)
agent-browser addinitscript <js>                  # Register at runtime (returns identifier)
agent-browser removeinitscript <identifier>       # Remove a previously registered init script

Setup

agent-browser install                 # Download Chrome from Chrome for Testing (Google's official automation channel)
agent-browser install --with-deps     # Also install system deps (Linux)
agent-browser upgrade                 # Upgrade agent-browser to the latest version
agent-browser doctor                  # Diagnose the install and auto-clean stale daemon files
agent-browser doctor --fix            # Also run destructive repairs (reinstall Chrome, purge old state, ...)
agent-browser doctor --offline --quick  # Skip network probes and the live launch test
agent-browser mcp

doctor checks your environment, Chrome install, daemon state, config files, encryption key, providers, network reachability, and runs a live headless browser launch test. Stale socket/pid sidecar files are auto-cleaned. Output is also available as --json for agents.

Skills

agent-browser skills                  # List available skills
agent-browser skills list             # Same as above
agent-browser skills get <name>       # Output a skill's full content
agent-browser skills get <name> --full  # Include references and templates
agent-browser skills get --all        # Output every skill
agent-browser skills path

Serves bundled skill content that always matches the installed CLI version. AI agents use this to get current instructions rather than relying on cached copies. Set AGENT_BROWSER_SKILLS_DIR to override the skills directory path.

MCP Server

agent-browser mcp
agent-browser mcp --tools all
agent-browser mcp --tools core,network,react

Starts a Model Context Protocol server over stdio. MCP clients launch this command as a subprocess and exchange newline-delimited JSON-RPC on stdin and stdout. The server defaults to MCP protocol 2025-11-25 and accepts older supported client protocol versions during initialization.

The default tools profile is core, which keeps MCP context small for everyday browser automation. Use --tools all for the full typed CLI parity surface, or combine profiles with commas, such as --tools core,network,react.

Profiles:

• core — Default. Navigation, snapshots, interaction, waits, reads, screenshots, JavaScript eval, close, tab basics, and profile discovery
• network — Network routes, request inspection, HAR, headers, credentials, offline
• state — Cookies, storage, auth, saved state, sessions, profiles, skills
• debug — Console/errors, tracing, profiling, recording, a11y audit, clipboard, plugins, doctor, dashboard, install, upgrade, chat, diff, batch, confirm/deny
• tabs — Back/forward/reload, tabs, windows, frames, dialogs
• react — React tree/inspect/renders/suspense, vitals, pushstate
• mobile — Viewport/device/geolocation/media, touch, swipe, mouse, keyboard
• all — Every MCP tool, including the full typed CLI parity surface

Common tools include:

• agent_browser_tools_profiles
• agent_browser_open
• agent_browser_snapshot
• agent_browser_click
• agent_browser_fill
• agent_browser_type
• agent_browser_press
• agent_browser_wait_for_selector
• agent_browser_screenshot
• agent_browser_get_url
• agent_browser_eval
• agent_browser_close

Each tool has typed fields such as url, selector, text, key, session, and allowedDomains, so MCP clients show meaningful approval prompts instead of raw command arrays. The common allowedDomains array maps to --allowed-domains and activates the same WebRTC containment and launch-mode restrictions. Each tool also accepts extraArgs for advanced CLI flags and exact CLI parity. Tool discovery is paginated and includes read-only/open-world annotations so modern MCP clients can load the large typed surface incrementally.

Example MCP client config:

{
  "mcpServers": {
    "agent-browser": {
      "command": "agent-browser",
      "args": ["mcp"]
    }
  }
}

Full parity MCP client config:

{
  "mcpServers": {
    "agent-browser": {
      "command": "agent-browser",
      "args": ["mcp", "--tools", "all"]
    }
  }
}

Tool invocations use the same config files and environment variables as the CLI. Use session in the tool arguments, or set AGENT_BROWSER_SESSION, to isolate browser state.

Authentication

agent-browser provides multiple ways to persist login sessions so you don’t re-authenticate every run.

Quick summary

Import auth from your browser

If you are already logged in to a site in Chrome, you can grab that auth state and reuse it:

# 1. Launch Chrome with remote debugging enabled
#    macOS:
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" --remote-debugging-port=9222
#    Or use --auto-connect to discover an already-running Chrome

# 2. Connect and save the authenticated state
agent-browser --auto-connect state save ./my-auth.json

# 3. Use the saved auth in future sessions
agent-browser --state ./my-auth.json open https://app.example.com/dashboard

# 4. Or use --restore for automatic persistence
SESSION=

Security notes:

• --remote-debugging-port exposes full browser control on localhost. Any local process can connect. Only use on trusted machines and close Chrome when done.
• State files contain session tokens in plaintext. Add them to .gitignore and delete when no longer needed. For encryption at rest, set AGENT_BROWSER_ENCRYPTION_KEY (see State Encryption).

For full details on login flows, OAuth, 2FA, cookie-based auth, and the auth vault, see the Authentication (opens in a new tab) docs.

Sessions

Run multiple isolated browser instances:

# Different sessions
agent-browser --session agent1 open site-a.com
agent-browser --session agent2 open site-b.com

# Or via environment variable
AGENT_BROWSER_SESSION=agent1 agent-browser click "#btn"

# List active sessions
agent-browser session list
# Output:
# Active sessions:
# -> default
#    agent1

Each session has its own:

• Browser instance
• Cookies and storage
• Navigation history
• Authentication state

Chrome Profile Reuse

The fastest way to use your existing login state: pass a Chrome profile name to --profile:

# List available Chrome profiles
agent-browser profiles

# Reuse your default Chrome profile's login state
agent-browser --profile Default open https://gmail.com

# Use a named profile (by display name or directory name)
agent-browser --profile "Work" open https://app.example.com

# Or via environment variable
AGENT_BROWSER_PROFILE=Default agent-browser open https://gmail.com

This copies your Chrome profile to a temp directory (read-only snapshot, no changes to your original profile), so the browser launches with your existing cookies and sessions.

Note: On Windows, close Chrome before using --profile <name> if Chrome is running, as some profile files may be locked.

Persistent Profiles

For a persistent custom profile directory that stores state across browser restarts, pass a path to --profile:

# Use a persistent profile directory
agent-browser --profile ~/.myapp-profile open myapp.com

# Login once, then reuse the authenticated session
agent-browser --profile ~/.myapp-profile open myapp.com/dashboard

# Or via environment variable
AGENT_BROWSER_PROFILE=~/.myapp-profile agent-browser open myapp.com

The profile directory stores:

• Cookies and localStorage
• IndexedDB data
• Service workers
• Browser cache
• Login sessions

Tip: Use different profile paths for different projects to keep their browser state isolated.

Session Persistence

Use --restore with a stable --session to automatically save and restore cookies and localStorage across browser restarts:

# Generate a stable id for this worktree and auto-save/load state
SESSION="$(agent-browser session id --scope worktree --prefix twitter)"
agent-browser --session "$SESSION" --restore open twitter.com

# Login once, then state persists automatically
# State files stored in ~/.agent-browser/sessions/

# Optional: validate restored state before auto-saving again
agent-browser --session "$SESSION" --restore --restore-check-text

State is saved when the browser closes (explicit close, idle timeout, or daemon shutdown) and also periodically while the browser is open, so a browser window you close by hand still leaves a recent save behind. Periodic autosave waits for commands to settle, then saves at most once per AGENT_BROWSER_AUTOSAVE_INTERVAL_MS (default 30000; set to 0 to save only on close). Idle sessions keep saving on the same interval, so changes the page makes on its own (token refreshes, background requests) are captured too. It respects the --restore-save policy.

State Encryption

Encrypt saved session data at rest with AES-256-GCM:

# Generate key: openssl rand -hex 32
export AGENT_BROWSER_ENCRYPTION_KEY=<64-char-hex-key>

# State files are now encrypted automatically
agent-browser --session secure --restore open example.com

Security

agent-browser includes security features for safe AI agent deployments. All features are opt-in, and existing workflows are unaffected until you explicitly enable a feature:

• Authentication Vault: Store credentials locally (always encrypted), reference by name. The LLM never sees passwords. auth login navigates with load and then waits for login form selectors to appear (SPA-friendly, timeout follows the default action timeout). A key is auto-generated at ~/.agent-browser/.encryption-key if AGENT_BROWSER_ENCRYPTION_KEY is not set: echo "pass" | agent-browser auth save github --url https://github.com/login --username user --password-stdin then agent-browser auth login github
• Plugin System: Extend agent-browser with external executable plugins. Plugins run out-of-process over the agent-browser.plugin.v1 stdio JSON protocol and declare capabilities such as credential.read, browser.provider, launch.mutate, or command.run.
• Content Boundary Markers: Wrap page output in delimiters so LLMs can distinguish tool output from untrusted content: --content-boundaries
• Domain Allowlist: Restrict navigation to trusted domains (wildcards like *.example.com also match the bare domain): --allowed-domains "example.com,*.example.com". Sub-resource requests (scripts, images, fetch), WebSocket/EventSource connections, and sendBeacon calls to non-allowed domains are blocked. WebRTC peer connections are disabled in supported Chromium sessions while the allowlist is active to prevent STUN, TURN, and DNS traffic from bypassing HTTP interception. Dedicated and shared workers are guarded with a bootstrap wrapper; if a page CSP forbids that wrapper, the worker fails closed rather than running without the allowlist guard. Pre-existing CDP sessions, auto-connect, Chrome profiles, direct-page provider plugins, agent-browser restore or state-file replay, raw Chrome args that select profiles, restore sessions, or open startup pages, iOS, and Safari reject this option because agent-browser cannot install equivalent containment before page scripts run. Include any CDN domains your target pages depend on (e.g., *.cdn.example.com).
• Action Policy: Gate destructive actions with a static policy file: --action-policy ./policy.json
• Action Confirmation: Require explicit approval for sensitive action categories: --confirm-actions eval,download
• Output Length Limits: Prevent context flooding: --max-output 50000

See Security documentation (opens in a new tab) for details.

Plugin System

Plugins let third-party tools integrate without becoming built-in agent-browser dependencies. Add a plugin from npm or GitHub:

agent-browser plugin add agent-browser-plugin-captcha
agent-browser plugin add @company/agent-browser-plugin-vault --name vault
agent-browser plugin add org/agent-browser-plugin-cloud-browser

References are resolved by shape: name uses npm, @scope/name uses npm, and owner/repo uses GitHub. plugin add writes ./agent-browser.json by default; use --global for ~/.agent-browser/config.json.

Plugin packages should support plugin.manifest so plugin add can discover their name and capabilities automatically. If a plugin does not support manifests, pass --capability <name> during add.

Plugins can also be configured manually in agent-browser.json:

{
  "plugins": [
    {
      "name": "vault",
      "command": "agent-browser-plugin-vault",
      "capabilities": ["credential.read"]
    },
    {
      "name": "cloud-browser",
      "command": "agent-browser-plugin-cloud-browser",
      "capabilities": [

Inspect configured plugins:

agent-browser plugin list
agent-browser plugin show vault

Use a credential provider plugin for one login:

agent-browser auth login my-app --credential-provider vault --item "My App"
agent-browser auth login my-app --credential-provider vault --item "My App" --url https://app.example.com/login --username-selector "#email" --password-selector "#password" --submit-selector "button[type=submit]"

Use a browser provider plugin:

agent-browser --provider cloud-browser open https://example.com

Use a launch mutator plugin for stealth or local launch customization. The plugin can append Chrome args, extensions, and init scripts before the browser starts:

agent-browser open https://example.com

Use a generic plugin command for domain-specific tools such as CAPTCHA solvers:

agent-browser plugin run captcha captcha.solve --payload '{"siteKey":"...","url":"https://example.com"}'

The protocol request always includes protocol, type, capability, and request. A credential plugin receives credential.resolve, a browser provider receives browser.launch, a launch mutator receives launch.mutate, and generic commands receive the supplied request type. plugin run is for command.run and custom capabilities; core capabilities and protocol request types use their dedicated command paths. agent-browser keeps browser automation, redaction-sensitive output, and policy enforcement in core.

Gate plugin access by capability action:

agent-browser --confirm-actions plugin:vault:credential.read auth login my-app --credential-provider vault --item "My App"
agent-browser --confirm-actions plugin:cloud-browser:browser.provider --provider cloud-browser open https://example.com
agent-browser --confirm-actions plugin:stealth:launch.mutate open https://example.com

Do not put vault tokens or passwords in plugin command args. Use the vault vendor’s own login/session mechanism or environment outside agent-browser config.

Snapshot Options

The snapshot command supports filtering to reduce output size:

agent-browser snapshot                    # Full accessibility tree
agent-browser snapshot -i                 # Interactive elements only (buttons, inputs, links)
agent-browser snapshot -i --urls          # Interactive elements with link URLs
agent-browser snapshot -c                 # Compact (remove empty structural elements)
agent-browser snapshot -d 3               # Limit depth to 3 levels
agent-browser snapshot -s "#main"         # Scope to CSS selector
agent-browser

Annotated Screenshots

The --annotate flag overlays numbered labels on interactive elements in the screenshot. Each label [N] corresponds to ref @eN, so the same refs work for both visual and text-based workflows.

Annotated screenshots are supported on the CDP-backed browser path (Chrome/Lightpanda). The Safari/WebDriver backend does not yet support --annotate.

agent-browser screenshot --annotate
# -> Screenshot saved to /tmp/screenshot-2026-02-17T12-00-00-abc123.png
#    [1] @e1 button "Submit"
#    [2] @e2 link "Home"
#    [3] @e3 textbox "Email"

After an annotated screenshot, refs are cached so you can immediately interact with elements:

agent-browser screenshot --annotate ./page.png
agent-browser click @e2     # Click the "Home" link labeled [2]

This is useful for multimodal AI models that can reason about visual layout, unlabeled icon buttons, canvas elements, or visual state that the text accessibility tree cannot capture.

Options

Observability Dashboard

Monitor agent-browser sessions in real time with a local web dashboard showing a live viewport and command activity feed.

# Start the dashboard server (runs in background on port 4848)
agent-browser dashboard start
agent-browser dashboard start --port 8080   # Custom port

# All sessions are automatically visible in the dashboard
agent-browser open example.com

# Stop the dashboard
agent-browser dashboard stop

The dashboard runs as a standalone background process on port 4848, independent of browser sessions. It stays available even when no sessions are running, and it works from http://localhost:4848 or a proxied/forwarded URL that reaches the dashboard server, such as https://dashboard.agent-browser.localhost or a Coder workspace URL. The browser stays on the dashboard origin; session-specific tabs, status, and stream traffic are proxied internally, so session ports do not need to be exposed.

The dashboard displays:

• Live viewport: real-time JPEG frames from the browser
• Activity feed: chronological command/result stream with timing and expandable details
• Console output: browser console messages (log, warn, error)
• Session creation: create new sessions from the UI with local engines (Chrome, Lightpanda) or cloud providers (AgentCore, Browserbase, Browserless, Browser Use, Kernel)
• AI Chat: chat with an AI assistant directly in the dashboard (requires Vercel AI Gateway configuration)

AI Chat

The dashboard includes an optional AI chat panel powered by the Vercel AI Gateway. The same functionality is available directly from the CLI via the chat command. Set these environment variables to enable AI chat:

export AI_GATEWAY_API_KEY=gw_your_key_here
export AI_GATEWAY_MODEL=anthropic/claude-sonnet-4.6           # optional, this is the default
export AI_GATEWAY_URL=https://ai-gateway.vercel.sh           # optional, this is the default

CLI usage:

agent-browser chat "open google.com and search for cats"     # Single-shot
agent-browser chat                                           # Interactive REPL
agent-browser -q chat "summarize this page"                  # Quiet mode (text only)
agent-browser -v chat "fill in the login form"               # Verbose (show command output)
agent-browser --model openai/gpt-4o chat "take a screenshot" # Override model

The chat command translates natural language instructions into agent-browser commands, executes them, and streams the AI response. In interactive mode, type quit to exit. Use --json for structured output suitable for agent consumption.

Dashboard usage:

The Chat tab is always visible in the dashboard. When AI_GATEWAY_API_KEY is set, the Rust server proxies requests to the gateway and streams responses back using the Vercel AI SDK’s UI Message Stream protocol. Without the key, sending a message shows an error inline.

Configuration

Create an agent-browser.json file to set persistent defaults instead of repeating flags on every command.

Locations (lowest to highest priority):

1. ~/.agent-browser/config.json: user-level defaults
2. ./agent-browser.json: project-level overrides (in working directory)
3. AGENT_BROWSER_* environment variables override config file values
4. CLI flags override everything

Example agent-browser.json:

{
  "headed": true,
  "proxy": "http://localhost:8080",
  "profile": "./browser-data",
  "userAgent": "my-agent/1.0",
  "hideScrollbars": false,
  "ignoreHttpsErrors": true,
  "plugins": [
    {
      "name": "vault",

Use --config <path> or AGENT_BROWSER_CONFIG to load a specific config file instead of the defaults:

agent-browser --config ./ci-config.json open example.com
AGENT_BROWSER_CONFIG=./ci-config.json agent-browser open example.com

All options from the table above can be set in the config file using camelCase keys (e.g., --executable-path becomes "executablePath", --proxy-bypass becomes "proxyBypass"). Plugins are configured with the "plugins" array shown above. Unknown keys are ignored for forward compatibility.

A JSON Schema (opens in a new tab) is available for IDE autocomplete and validation. Add a $schema key to your config file to enable it:

{
  "$schema": "https://agent-browser.dev/schema.json",
  "headed": true
}

Boolean flags accept an optional true/false value to override config settings. For example, --headed false disables "headed": true from config. A bare --headed is equivalent to --headed true.

Auto-discovered config files that are missing are silently ignored. If --config <path> points to a missing or invalid file, agent-browser exits with an error. Extensions from user and project configs are merged (concatenated), not replaced.

Tip: If your project-level agent-browser.json contains environment-specific values (paths, proxies), consider adding it to .gitignore.

Default Timeout

The default timeout for standard operations (clicks, waits, fills, etc.) is 25 seconds. This is intentionally below the CLI’s 30-second IPC read timeout so that the daemon returns a proper error instead of the CLI timing out with EAGAIN.

Override the default timeout via environment variable:

# Set a longer timeout for slow pages (in milliseconds)
export AGENT_BROWSER_DEFAULT_TIMEOUT=45000

Note: Setting this above 30000 (30s) may cause EAGAIN errors on slow operations because the CLI’s read timeout will expire before the daemon responds. The CLI retries transient errors automatically, but response times will increase.

Selectors

Refs (Recommended for AI)

Refs provide deterministic element selection from snapshots:

# 1. Get snapshot with refs
agent-browser snapshot
# Output:
# - heading "Example Domain" [ref=e1] [level=1]
# - button "Submit" [ref=e2]
# - textbox "Email" [ref=e3]
# - link "Learn more" [ref=e4]

# 2. Use refs to interact
agent-browser click @e2                   # Click the button
agent-browser fill @e3 "test@example.com" # Fill the textbox
agent-browser get text @e1                # Get heading text
agent-browser hover

When a ref click is blocked by an overlay, the error includes the covering element, such as covered by <div#consent-banner>. Click the banner or dialog control first, then run snapshot again before reusing refs.

Why use refs?

• Deterministic: Ref points to exact element from snapshot
• Fast: No DOM re-query needed
• AI-friendly: Snapshot + ref workflow is optimal for LLMs

CSS Selectors

agent-browser click "#id"
agent-browser click ".class"
agent-browser click "div > button"

Text & XPath

agent-browser click "text=Submit"
agent-browser click "xpath=//button"

Semantic Locators

agent-browser find role button click --name "Submit"
agent-browser find label "Email" fill "test@test.com"

Agent Mode

Use --json for machine-readable output:

agent-browser snapshot --json
# Returns: {"success":true,"data":{"snapshot":"...","refs":{"e1":{"role":"heading","name":"Title"},...}}}

agent-browser get text @e1 --json
agent-browser is visible @e2 --json

Optimal AI Workflow

# 1. Navigate and get snapshot
agent-browser open example.com
agent-browser snapshot -i --json   # AI parses tree and refs

# 2. AI identifies target refs from snapshot
# 3. Execute actions using refs
agent-browser click @e2
agent-browser fill @e3 "input text"

# 4. Get new snapshot if page changed
agent-browser snapshot -i --json

Command Chaining

Commands can be chained with && in a single shell invocation. The browser persists via a background daemon, so chaining is safe and more efficient:

# Open, wait for load, and snapshot in one call
agent-browser open example.com && agent-browser wait --load networkidle && agent-browser snapshot -i

# Chain multiple interactions
agent-browser fill @e1 "user@example.com" && agent-browser fill @e2 "pass" && agent-browser click @e3

# Navigate and screenshot
agent-browser open

Use && when you don’t need intermediate output. Run commands separately when you need to parse output first (e.g., snapshot to discover refs before interacting).

Headed Mode

Show the browser window for debugging:

agent-browser open example.com --headed

This opens a visible browser window instead of running headless.

On Linux hosts with no display (servers, containers), --headed still works: when DISPLAY is unset and Xvfb is installed, agent-browser starts a private virtual display for the browser and cleans it up on close (opt out with AGENT_BROWSER_NO_XVFB=1). Needed for WebGPU screenshots, and useful for extensions that misbehave headless.

Note: Browser extensions work in both headed and headless mode (Chrome’s --headless=new).

WebGPU

Headless Chrome does not expose WebGPU by default, so pages using it (three.js WebGPURenderer, Babylon.js, etc.) silently render black. The --webgpu flag enables a launch preset that makes WebGPU work, including in GPU-less containers and CI:

agent-browser --webgpu open https://my-webgpu-app.example.com
agent-browser screenshot app.png

On macOS and Windows this uses the hardware Metal/D3D backend. On Linux it routes WebGPU through SwiftShader’s software Vulkan (no GPU needed), which requires the system Vulkan loader and Mesa ICD:

apt-get install -y libvulkan1 mesa-vulkan-drivers

One upstream caveat: headless Chrome cannot capture WebGPU canvas presentation in screenshots on Windows and Linux (rendering and in-page readbacks work; the capture is black). Screenshots of WebGPU pages work headless on macOS; on Windows run --headed in a logged-in desktop session; on Linux just add --headed — when no DISPLAY is set and Xvfb is installed, agent-browser starts a private virtual display automatically (opt out with AGENT_BROWSER_NO_XVFB=1).

Verify the full pipeline (adapter, render pass, and screenshot capture) with:

agent-browser doctor --webgpu

Notes for WebGPU pages:

• WebGPU only exists in secure contexts (https://, http://localhost, or file://).
• three.js WebGPURenderer initializes asynchronously and silently falls back to WebGL2 when no adapter is available — wait for the app to render its first frame before taking a screenshot.
• To prefer a real GPU on Linux instead of SwiftShader, override both the Vulkan driver and the adapter with --args "--use-vulkan=native,--use-webgpu-adapter=default" (user args win over the preset; --use-webgpu-adapter alone still enumerates only SwiftShader).

See the WebGPU docs page (opens in a new tab) for the full platform matrix and container recipe.

Authenticated Sessions

Use --headers to set HTTP headers for a specific origin, enabling authentication without login flows:

# Headers are scoped to api.example.com only
agent-browser open api.example.com --headers '{"Authorization": "Bearer <token>"}'

# Requests to api.example.com include the auth header
agent-browser snapshot -i --json
agent-browser click @e2

# Navigate to another domain - headers are NOT sent (safe!)
agent-browser open other-site.com

This is useful for:

• Skipping login flows - Authenticate via headers instead of UI
• Switching users - Start new sessions with different auth tokens
• API testing - Access protected endpoints directly
• Security - Headers are scoped to the origin, not leaked to other domains

To set headers for multiple origins, use --headers with each open command:

agent-browser open api.example.com --headers '{"Authorization": "Bearer token1"}'
agent-browser open api.acme.com --headers '{"Authorization": "Bearer token2"}'

For global headers (all domains), use set headers:

agent-browser set headers '{"X-Custom-Header": "value"}'

Custom Browser Executable

Use a custom browser executable instead of the bundled Chromium. This is useful for:

• Serverless deployment: Use lightweight Chromium builds like @sparticuz/chromium (~50MB vs ~684MB)
• System browsers: Use an existing Chrome/Chromium installation
• Custom builds: Use modified browser builds

CLI Usage

# Via flag
agent-browser --executable-path /path/to/chromium open example.com

# Via environment variable
AGENT_BROWSER_EXECUTABLE_PATH=/path/to/chromium agent-browser open example.com

Serverless (Vercel)

Run agent-browser + Chrome in an ephemeral Vercel Sandbox microVM. No external server needed:

import { runAgentBrowserCommand, withAgentBrowserSandbox } from "@agent-browser/sandbox/vercel";

const result = await withAgentBrowserSandbox(async (sandbox) => {
  await runAgentBrowserCommand(sandbox, ["open", "https://example.com"]);
  return runAgentBrowserCommand(sandbox, ["screenshot"]);
});

Install @agent-browser/sandbox and @vercel/sandbox in the consuming app. See the sandbox helper example (opens in a new tab) for minimal Vercel Sandbox usage, or the environments example (opens in a new tab) for a full UI demo with a deploy-to-Vercel button.

Fresh Vercel and eve sandboxes install Chromium system dependencies by default. Pass installSystemDependencies: false only when your sandbox image already includes those libraries.

eve extension

Give an eve (opens in a new tab) agent the full browser tool set by mounting the @agent-browser/eve (opens in a new tab) extension:

// agent/extensions/browser.ts
import browser from "@agent-browser/eve";

export default browser({});

This composes ~20 namespaced tools into the agent — browser__navigate, browser__snapshot, browser__click, browser__fill, browser__find, browser__screenshot, and more — all running agent-browser inside the agent’s sandbox. agent-browser installs automatically on first use; pre-install it in agent/sandbox.ts with the @agent-browser/eve/sandbox helpers to bake the cost into the sandbox template instead. Configuration (domain allowlists, output limits, session naming) and per-tool overrides are covered in the package README (opens in a new tab), and the eve example (opens in a new tab) is a complete app with the extension mounted.

Serverless (AWS Lambda)

import chromium from '@sparticuz/chromium';
import { execSync } from 'child_process';

export async function handler() {
  const executablePath = await chromium.executablePath();
  const result = execSync(
    `AGENT_BROWSER_EXECUTABLE_PATH=${executablePath} agent-browser open https://example.com && agent-browser snapshot -i --json`,

Local Files

Open and interact with local files (PDFs, HTML, etc.) using file:// URLs:

# Enable file access (required for JavaScript to access local files)
agent-browser --allow-file-access open file:///path/to/document.pdf
agent-browser --allow-file-access open file:///path/to/page.html

# Take screenshot of a local PDF
agent-browser --allow-file-access open file:///Users/me/report.pdf
agent-browser screenshot report.png

The --allow-file-access flag adds Chromium flags (--allow-file-access-from-files, --allow-file-access) that allow file:// URLs to:

• Load and render local files
• Access other local files via JavaScript (XHR, fetch)
• Load local resources (images, scripts, stylesheets)

Note: This flag only works with Chromium. For security, it’s disabled by default.

CDP Mode

Connect to an existing browser via Chrome DevTools Protocol:

# Start Chrome with: google-chrome --remote-debugging-port=9222

# Connect once, then run commands without --cdp
agent-browser connect 9222
agent-browser snapshot
agent-browser tab
agent-browser close

# Or pass --cdp on each command
agent-browser --cdp 9222 snapshot

# Connect to remote browser via WebSocket URL
agent-browser --cdp "wss://your-browser-service.com/cdp?token=..." snapshot

The --cdp flag accepts either:

• A port number (e.g., 9222) for local connections via http://localhost:{port}
• A full WebSocket URL (e.g., wss://... or ws://...) for remote browser services

This enables control of:

• Electron apps
• Chrome/Chromium instances with remote debugging
• WebView2 applications
• Any browser exposing a CDP endpoint

Auto-Connect

Use --auto-connect to automatically discover and connect to a running Chrome instance without specifying a port:

# Auto-discover running Chrome with remote debugging
agent-browser --auto-connect open example.com
agent-browser --auto-connect snapshot

# Or via environment variable
AGENT_BROWSER_AUTO_CONNECT=1 agent-browser snapshot

Auto-connect discovers Chrome by:

1. Reading Chrome’s DevToolsActivePort file from the default user data directory
2. Falling back to probing common debugging ports (9222, 9229)
3. If HTTP-based discovery (/json/version, /json/list) fails, falling back to a direct WebSocket connection

This is useful when:

• Chrome 144+ has remote debugging enabled via chrome://inspect/#remote-debugging (which uses a dynamic port)
• You want a zero-configuration connection to your existing browser
• You don’t want to track which port Chrome is using

Streaming (Browser Preview)

Stream the browser viewport via WebSocket for live preview or “pair browsing” where a human can watch and interact alongside an AI agent.

Streaming

Every session automatically starts a WebSocket stream server on an OS-assigned port. Use stream status to see the bound port and connection state:

agent-browser stream status

To bind to a specific port, set AGENT_BROWSER_STREAM_PORT:

AGENT_BROWSER_STREAM_PORT=9223 agent-browser open example.com

Frame encoding is daemon-wide:

Width and height cap the encoded frame and leave the page size alone, so a portrait or HiDPI viewport keeps its resolution unless you cap it. The live stream requests jpeg. An explicit screencast_start reconfigures the same screencast, so a client can see the format change mid-stream. On a busy page at 1280x720, quality 80 costs about 54 KB per frame, quality 20 about 25 KB, and quality 20 at 640x360 about 9 KB.

# Cheaper frames for a constrained link
AGENT_BROWSER_STREAM_QUALITY=20 \
AGENT_BROWSER_STREAM_MAX_WIDTH=640 \
AGENT_BROWSER_STREAM_MAX_HEIGHT=360 \
agent-browser open example.com

You can also manage streaming at runtime with stream enable, stream disable, and stream status:

agent-browser stream enable --port 9223   # Re-enable on a specific port
agent-browser stream disable              # Stop streaming for the session

The WebSocket server streams the browser viewport and accepts input events.

WebSocket Protocol

Connect to ws://localhost:9223 to receive frames and send input:

Receive frames:

{
  "type": "frame",
  "seq": 41,
  "data": "<base64-encoded-jpeg>",
  "metadata": {
    "deviceWidth": 1280,
    "deviceHeight": 720,
    "pageScaleFactor": 1,
    "offsetTop": 0,
    "scrollOffsetX": 

seq is a monotonic frame id, echoed back in an ack message under ack pacing. metadata.timestamp is the capture time in epoch milliseconds, so a client can tell how old a frame is by the time it draws it.

Send mouse events:

{
  "type": "input_mouse",
  "eventType": "mousePressed",
  "x": 100,
  "y": 200,
  "button": "left",
  "clickCount": 1
}

Send keyboard events:

{
  "type": "input_keyboard",
  "eventType": "keyDown",
  "key": "Enter",
  "code": "Enter"
}

Send touch events:

{
  "type": "input_touch",
  "eventType": "touchStart",
  "touchPoints": [{ "x": 100, "y": 200 }]
}

Cap the frame rate (per client):

{
  "type": "config",
  "maxFps": 10
}

Frames are delivered latest-first: the server picks the newest frame at send time, so frames produced while an earlier one is still being written are skipped rather than queued. maxFps (1 to 120, 0 = uncapped) limits delivery for that client only. A client that sends {"type":"config","pacing":"ack"} receives one frame at a time and acknowledges it with {"type":"ack","seq":N}, so nothing stale reaches the socket even if that client stalls; in the default push pacing, frames already handed to the transport are still delivered in order. Both settings can also be declared on the URL (ws://127.0.0.1:<port>/?pacing=ack&maxFps=10), which is the only way to cover the connection’s opening frame. Input events are read on a dedicated task per connection, so clicks and keystrokes dispatch immediately even while frames are mid-write to a slow client. They are sent to the browser without waiting for its reply, so a click stays responsive behind a burst of mouse moves, and ordering is preserved.

Architecture

agent-browser uses a client-daemon architecture:

1. Rust CLI - Parses commands, communicates with daemon
2. Rust Daemon - Pure Rust daemon using direct CDP, no Node.js required

The daemon starts automatically on first command and persists between commands for fast subsequent operations. After 1 hour with no commands or dashboard input it saves configured restore state, closes the browser, and exits, so an integration that dies without calling close cannot leak the daemon and its browser indefinitely; the next command starts a fresh daemon and configured state restore works as usual. A session without --restore or another restore key does not save browser state, so its transient state and open tabs are discarded at shutdown. Set --idle-timeout to a duration such as 30s, 5m, or 1h, or set AGENT_BROWSER_IDLE_TIMEOUT_MS to a value in milliseconds. Use 0 to disable idle shutdown entirely. The default never closes a headed browser, including Safari and iOS WebDriver sessions, or a user-attached browser because those may be in direct human use. Provider-owned cloud browsers remain eligible for cleanup. An explicitly set timeout applies to every browser.

Browser Engine: Uses Chrome (from Chrome for Testing) by default. The --engine flag selects between chrome and lightpanda. Supported browsers: Chromium/Chrome (via CDP) and Safari (via WebDriver for iOS).

Platforms

Usage with AI Agents

Just ask the agent

The simplest approach is to tell your agent to use it:

Use agent-browser to test the login flow. Run agent-browser --help to see available commands.

The --help output is comprehensive and most agents can figure it out from there.

AI Coding Assistants (recommended)

Add the skill to your AI coding assistant for richer context:

npx skills add vercel-labs/agent-browser

This works with Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Goose, OpenCode, and Windsurf. The skill is fetched from the repository, so it stays up to date automatically. Do not copy SKILL.md from node_modules as it will become stale.

Claude Code

Install as a Claude Code skill:

npx skills add vercel-labs/agent-browser

This adds a thin discovery stub at .claude/skills/agent-browser/SKILL.md. The stub is intentionally minimal — it points Claude Code at agent-browser skills get core to load the actual workflow content at runtime. This way the instructions always match the installed CLI version instead of going stale between releases.

AGENTS.md / CLAUDE.md

For more consistent results, add to your project or global instructions file:

## Browser Automation

Use `agent-browser` for web automation. Run `agent-browser --help` for all commands.

Core workflow:

1. `agent-browser open <url>` - Navigate to page
2. `agent-browser snapshot -i` - Get interactive elements with refs (@e1, @e2)
3. `agent-browser click @e1` / `fill @e2 "text"` - Interact using refs
4. Re-snapshot after page changes

Integrations

iOS Simulator

Control real Mobile Safari in the iOS Simulator for authentic mobile web testing. Requires macOS with Xcode.

Setup:

# Install Appium and XCUITest driver
npm install -g appium
appium driver install xcuitest

Usage:

# List available iOS simulators
agent-browser device list

# Launch Safari on a specific device
agent-browser -p ios --device "iPhone 16 Pro" open https://example.com

# Same commands as desktop
agent-browser -p ios snapshot -i
agent-browser -p ios tap @e1
agent-browser -p ios fill @e2 "text"

Or use environment variables:

export AGENT_BROWSER_PROVIDER=ios
export AGENT_BROWSER_IOS_DEVICE="iPhone 16 Pro"
agent-browser open https://example.com

Supported devices: All iOS Simulators available in Xcode (iPhones, iPads), plus real iOS devices.

Note: The iOS provider boots the simulator, starts Appium, and controls Safari. First launch takes ~30-60 seconds; subsequent commands are fast.

Real Device Support

Appium also supports real iOS devices connected via USB. This requires additional one-time setup:

1. Get your device UDID:

xcrun xctrace list devices
# or
system_profiler SPUSBDataType | grep -A 5 "iPhone\|iPad"

2. Sign WebDriverAgent (one-time):

# Open the WebDriverAgent Xcode project
cd ~/.appium/node_modules/appium-xcuitest-driver/node_modules/appium-webdriveragent
open WebDriverAgent.xcodeproj

In Xcode:

• Select the WebDriverAgentRunner target
• Go to Signing & Capabilities
• Select your Team (requires Apple Developer account, free tier works)
• Let Xcode manage signing automatically

3. Use with agent-browser:

# Connect device via USB, then:
agent-browser -p ios --device "<DEVICE_UDID>" open https://example.com

# Or use the device name if unique
agent-browser -p ios --device "John's iPhone" open https://example.com

Real device notes:

• First run installs WebDriverAgent to the device (may require Trust prompt)
• Device must be unlocked and connected via USB
• Slightly slower initial connection than simulator
• Tests against real Safari performance and behavior

Browserless

Browserless (opens in a new tab) provides cloud browser infrastructure with a Sessions API. Use it when running agent-browser in environments where a local browser isn’t available.

To enable Browserless, use the -p flag:

export BROWSERLESS_API_KEY="your-api-token"
agent-browser -p browserless open https://example.com

Or use environment variables for CI/scripts:

export AGENT_BROWSER_PROVIDER=browserless
export BROWSERLESS_API_KEY="your-api-token"
agent-browser open https://example.com

Optional configuration via environment variables:

When enabled, agent-browser connects to a Browserless cloud session instead of launching a local browser. All commands work identically.

Get your API token from the Browserless Dashboard (opens in a new tab).

Browserbase

Browserbase (opens in a new tab) provides remote browser infrastructure to make deployment of agentic browsing agents easy. Use it when running the agent-browser CLI in an environment where a local browser isn’t feasible.

To enable Browserbase, use the -p flag:

export BROWSERBASE_API_KEY="your-api-key"
agent-browser -p browserbase open https://example.com

Or use environment variables for CI/scripts:

export AGENT_BROWSER_PROVIDER=browserbase
export BROWSERBASE_API_KEY="your-api-key"
agent-browser open https://example.com

When enabled, agent-browser connects to a Browserbase session instead of launching a local browser. All commands work identically.

Get your API key from the Browserbase Dashboard (opens in a new tab).

Browser Use

Browser Use (opens in a new tab) provides cloud browser infrastructure for AI agents. Use it when running agent-browser in environments where a local browser isn’t available (serverless, CI/CD, etc.).

To enable Browser Use, use the -p flag:

export BROWSER_USE_API_KEY="your-api-key"
agent-browser -p browseruse open https://example.com

Or use environment variables for CI/scripts:

export AGENT_BROWSER_PROVIDER=browseruse
export BROWSER_USE_API_KEY="your-api-key"
agent-browser open https://example.com

When enabled, agent-browser connects to a Browser Use cloud session instead of launching a local browser. All commands work identically.

Get your API key from the Browser Use Cloud Dashboard (opens in a new tab). Free credits are available to get started, with pay-as-you-go pricing after.

Kernel

Kernel (opens in a new tab) provides cloud browser infrastructure for AI agents with features like stealth mode and persistent profiles.

To enable Kernel, use the -p flag:

export KERNEL_API_KEY="your-api-key"
agent-browser -p kernel open https://example.com

Or use environment variables for CI/scripts:

export AGENT_BROWSER_PROVIDER=kernel
export KERNEL_API_KEY="your-api-key"
agent-browser open https://example.com

Optional configuration via environment variables:

When enabled, agent-browser connects to a Kernel cloud session instead of launching a local browser. All commands work identically.

Profile Persistence: When KERNEL_PROFILE_NAME is set, the profile will be created if it doesn’t already exist. Cookies, logins, and session data are automatically saved back to the profile when the browser session ends, making them available for future sessions.

Get your API key from the Kernel Dashboard (opens in a new tab).

AgentCore

AWS Bedrock AgentCore (opens in a new tab) provides cloud browser sessions with SigV4 authentication.

To enable AgentCore, use the -p flag:

agent-browser -p agentcore open https://example.com

Or use environment variables for CI/scripts:

export AGENT_BROWSER_PROVIDER=agentcore
agent-browser open https://example.com

Credentials are automatically resolved from environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY) or the AWS CLI (aws configure export-credentials), which supports SSO, profiles, and IAM roles.

Optional configuration via environment variables:

Browser profiles: When AGENTCORE_PROFILE_ID is set, browser state (cookies, localStorage) is persisted across sessions automatically.

When enabled, agent-browser connects to an AgentCore cloud browser session instead of launching a local browser. All commands work identically.

License

Apache-2.0