AI ke haath mein tera Chrome.
Live. Remote. Secure.

Install one extension. Share a 9-digit code. Claude / Cowork / Cursor tera Chrome chala lega — typing, clicking, scraping, signups — live screencast ke saath, har action ke liye consent.

Get extension →See how it works
editpad.org
🤖 AI is typing · session #7341 · live
flow

How it works

5 steps, about 60 seconds. Zero remote-desktop software. Full per-action consent.

01
Install extension
User loads Browser Pilot in Chrome. 30 seconds, one-time.
02
Share 9-digit ID
Popup shows AnyDesk-style ID like 614 014 295. User sends it to the controller.
03
AI requests session
Controller calls bp_connect. The user sees an Accept / Reject prompt.
04
User accepts
Active session badge turns ON. 30-min TTL by default. User can end anytime.
05
AI works remotely
Forms, signups, scraping, anything. Live screencast for audit.
step 1 · user side

Install the Chrome extension

Currently an unlisted developer build. Chrome Web Store version coming soon.

  1. 01Get the Browser Pilot extension from Telegram @abhibots or GitHub
  2. 02Extract it to a clean folder on your machine
  3. 03Open chrome://extensions in Chrome
  4. 04Toggle Developer mode ON (top-right)
  5. 05Click Load unpacked and select the extracted folder
  6. 06Pin "Browser Pilot" from the puzzle-piece icon
  7. 07Click the icon — you'll see a 9-digit ID like 614 014 295
step 2 · controller side

MCP / API setup

Use it from Claude Desktop, Cowork, Cursor, or any MCP-compatible client. Plain HTTP / curl also works.

option A · MCP (Claude Desktop / Cowork)

Add to ~/Library/Application Support/Claude/claude_desktop_config.json, then restart Claude — bp_* tools become available.

{
  "mcpServers": {
    "browser-pilot": {
      "command": "node",
      "args": ["/path/to/browser-pilot/mcp-bridge/server.js"],
      "env": {
        "PILOT_HUB_URL": "https://pilot.abhibots.com",
        "PILOT_CONTROLLER_TOKEN": "your-token",
        "PILOT_CONTROLLER_NAME": "Your Name"
      }
    }
  }
}
option B · direct HTTP (curl / Python / any agent)

The hub is a plain REST & WebSocket API. All tools flow through /cmd.

# 1 — Hub health
curl https://pilot.abhibots.com/health

# 2 — Request session (user gets an Accept / Reject popup)
curl -X POST https://pilot.abhibots.com/session/request \
  -H "x-controller-token: $TOKEN" \
  -H "content-type: application/json" \
  -d '{"agent_id":"614014295","controller":"You","ttl_ms":1800000}'

# 3 — Send any command
curl -X POST https://pilot.abhibots.com/cmd \
  -H "x-controller-token: $TOKEN" \
  -d '{"session_id":"...","cmd":"tabs.create","params":{"url":"https://example.com"}}'
option C · live viewer (manual screen + mouse)

Open in any browser. Full live screencast plus remote mouse and keyboard.

https://pilot.abhibots.com/view?session_id=XXX&token=YYY
step 3 · capabilities

71+ tools — full Claude parity plus remote, screencast, and GIF

All tools are generic, not site-specific. The AI chains them based on the use case.

Connection
Session lifecycle + viewer URL
  • bp_status
    Hub health + connected agents
  • bp_connect
    Request session (60s for Accept)
  • bp_disconnect
    End session
  • bp_view_url
    Get live screencast viewer URL
Tabs & navigation
Tab control, URL navigation, window sizing
  • bp_tabs_list
    List all tabs
  • bp_tabs_create
    Open new tab
  • bp_tabs_activate
    Focus tab
  • bp_navigate
    URL / back / forward
  • bp_window_resize
    Resize window
  • bp_viewport
    Size, dpr, url, title
Mouse (real CDP)
Works on canvas, custom widgets, anything
  • bp_mouse_click
    Click at x,y
  • bp_double_click / bp_triple_click
    Double / triple click
  • bp_hover / bp_mouse_move
    Move cursor
  • bp_mouse_scroll
    Wheel scroll at x,y
  • bp_mouse_drag
    Drag (x1,y1) → (x2,y2)
Keyboard
Real OS-level key events + shortcuts
  • bp_keyboard_type
    Type text (unicode + emoji + Hindi)
  • bp_keyboard_press
    Enter, Tab, ArrowDown · combos cmd+a / ctrl+shift+t
Page interaction
Selector-based + smart find + content extraction
  • bp_click / bp_fill
    By CSS selector
  • bp_eval
    Run JS in MAIN world
  • bp_text / bp_page_readable
    Plain text + headings + links
  • bp_page_markdown
    Clean markdown
  • bp_page_snapshot
    Interactive elements + coords
  • bp_find
    Find by text query
  • bp_file_upload
    Base64 → <input type=file>
Accessibility tree (Claude-style)
Stable ref IDs for robust clicks
  • bp_ax_tree
    All interactive elements with role+name+coords
  • bp_ax_click
    Click by ref ID
  • bp_ax_scroll_to
    Scroll ref into view
Capture & recording
Screenshots, regions, animated GIFs (server-encoded)
  • bp_screenshot
    Full tab PNG
  • bp_screenshot_region
    Region / zoom capture
  • bp_gif_start / stop / export
    Record → ffmpeg encode → URL
Observation
Console, network, cookies — full devtools-like access
  • bp_console_read / clear
    console.log/warn/error buffer
  • bp_network_read / clear
    Network requests with filter
  • bp_cookies_get
    Cookies for URL
Audio · TTS · alerts
Speak / beep when AI needs human attention
  • bp_beep
    done / attention / error / intervention sounds
  • bp_speak
    TTS via OS voices
  • bp_alert
    Notification + beep + optional TTS combined
Orchestration
Atomic multi-step automation
  • bp_batch
    Array of {cmd, params, delay_ms} — single round-trip
Smart waits
Stop guessing sleep durations — wait for real signals
  • bp_wait_for_selector
    Visible / hidden / attached
  • bp_wait_for_text
    Until text appears in body
  • bp_wait_for_url
    Substring or regex URL match
  • bp_wait_for_network_idle
    No requests for idle_ms
  • bp_wait_sleep
    Fixed ms sleep
  • bp_wait_jitter
    Random delay — look human
Persistent state
Save + restore logged-in sessions across runs
  • bp_cookies_export
    Dump cookies for URL/domain
  • bp_cookies_import
    Restore saved cookies
  • bp_storage_export
    localStorage + sessionStorage
  • bp_storage_import
    Restore storage on origin
  • bp_state_snapshot
    One-shot full session capture
CAPTCHA & 2FA
Detect challenges, ping owner, autofill TOTP
  • bp_captcha_detect
    reCAPTCHA / hCaptcha / Turnstile / Arkose
  • bp_captcha_watch_start
    Background watcher → alert owner
  • bp_captcha_watch_stop
    Stop watcher
  • bp_totp
    6-digit code from base32 secret
Downloads
Trigger, track, fetch file path
  • bp_download_start
    By URL or click selector
  • bp_download_wait
    Wait until complete → file path
  • bp_download_cancel
    Abort in-progress
  • bp_downloads_recent
    List recent with state + size
Input lock per-tab
Block human cursor while automation runs
  • bp_lock
    Lock mouse + keyboard on tab
  • bp_unlock
    Release lock
Macros · ready-made flows
One-call recipes for Google, Gmail, LinkedIn, generic forms
  • bp_google_search
    SERP snippets (title+url+desc)
  • bp_google_open_first
    Search + open #1 organic result
  • bp_gmail_send
    Compose + send via web UI
  • bp_gmail_inbox
    Read inbox summary
  • bp_linkedin_open
    Open profile by name or URL
  • bp_linkedin_profile_read
    Structured profile data
  • bp_linkedin_connect
    Send connect with note
  • bp_linkedin_connect_batch
    Bulk-connect with human jitter
  • bp_form_detect
    Detect fields + labels + types
  • bp_form_fill
    Fill from {label: value} map
Audit & webhooks
Replay sessions + push events to your backend
  • bp_audit_get
    Fetch command log since index
  • bp_webhook_register
    Subscribe to session events
  • bp_webhook_unregister
    Unsubscribe
in action

Example: natural language → tool chain

The AI decides which tool to invoke. All steps are atomic and inspectable.

//user saysConnect to 614 014 295, open editpad.org, type 'Hello AI', beep when done.
1bp_connect"614014295"
↳userpopup → Accept
2bp_tabs_create"https://www.editpad.org/"
3bp_find"textarea" → (700, 431)
4bp_mouse_clicktabId, 700, 431
5bp_keyboard_type"Hello AI"
6bp_beep{type:"done"} → 🔔
who uses this

Use cases

Repetitive browser work that humans currently do — automate all of it.

· scale
Service business

Run LinkedIn outreach, Instagram DMs, signups across your client browsers — no AnyDesk needed. One AI handles 50+ users in parallel.

· support
Support / agency

Help users fill government portals, refunds, KYC. They install once, you assist remotely from any terminal.

· personal
Personal automation

Tell your AI "check attendance on these 5 college portals" and it runs in the background on your own sessions.

· dev
Developer / agent builder

Pure MCP API for your own AI agents. No vendor lock-in. Self-host the hub. Full source coming.

trust

Security model

Consent-first. Visible. Auditable. AnyDesk-style permissions — but only for the AI you grant access to.

Per-action Accept
Every new session requires the user to press Accept in the extension popup. No bypass.
Session TTL
30 minutes by default and configurable. Sessions auto-expire. The user can End at any time.
Visible badge
An ON badge appears on the extension toolbar icon whenever a session is active.
Chrome banner
Chrome shows its native "Browser Pilot started debugging this browser" banner. It cannot be hidden.
Controller token
Controllers authenticate to the hub with a private token. Without it, /cmd and /session/request return 401.
Audit logs
All commands are logged server-side. Full trail of who did what and when.
tech

Architecture

A boring stack. Node + WebSockets + Chrome MV3. No magic. Fully self-hostable.

┌──────────────────────────────────┐
│  Controller (AI / MCP / curl)    │
└────────────────┬─────────────────┘
                 │ HTTPS  /cmd + controller token
                 ▼
┌──────────────────────────────────┐
│  Hub @ pilot.abhibots.com        │  ← persistent WSS from each agent
│  · session routing               │
│  · screencast forwarding         │
│  · audit log                     │
└────────────────┬─────────────────┘
                 │ WSS  routed by 9-digit agent id
                 ▼
┌──────────────────────────────────┐
│  Chrome Extension MV3            │
│  · chrome.debugger (CDP)         │
│  · chrome.scripting              │
│  · offscreen audio + TTS         │
└────────────────┬─────────────────┘
                 ▼
        User's browser tab
vs

Browser Pilot vs Claude's Chrome extension

Claude's extension is solid — but local-only. If you need cross-machine control, that's the gap Browser Pilot fills.

capability
Claude Chrome
Browser Pilot
Remote control (cross-machine)
NoLocal only
YesAnywhere
Multi-user / multi-agent
No
YesMany agents on one hub
Per-session consent flow
Device pairing
YesPer-action Accept
Live screen viewer + remote mouse
No
YesCDP screencast + input
Open / self-hostable
No
YesComing
Hindi / regional language UX
No
Yes
Pricing flexibility
Anthropic-owned
YesFree / your hub