AI Browser Bridge

Lets any local AI (Claude Code, curl, scripts) drive your real Chrome tabs — click, type, fill forms, read page content, screenshot, navigate.

Sandbox (zero install)

https://fitzyracing1.github.io/nexus-ai/sandbox/ — a browser inside the webpage, with the bridge "extension" built in. Nothing to install: the window has an address bar, tabs, back/forward, and a set of real demo sites, and an AI console drives them for real — clicking buttons, filling forms, reading text, navigating, screenshotting. There's also an autonomous-task runner that chains commands ("buy the headphones", "fill & submit the sign-in form").

It works because the sandbox pages are served from the same origin, so in-page JavaScript can script them directly. The agent (sandbox/agent.js) is the same action logic that ships in extension/content.js — the same text= / text~= / CSS selector engine, the same synthetic mouse/keyboard events, the same controlled-input setter. The real extension does the identical thing to any tab; it just reaches it through the local bridge instead of an iframe. Use the sandbox to see exactly what the bridge can do before installing.

Live demo (real bridge)

https://fitzyracing1.github.io/nexus-ai/ — a real, working demo (not a mock). Open it in an extensions-capable Chromium browser (Chrome, Edge, Brave) and it drives that very page through your local bridge:

1. Start bridge.py and install the extension (see setup).
2. Open the demo page and paste your bridge auth token (from http://127.0.0.1:17777/setup) into the token box.
3. The three status boxes should turn green (browser ✓, extension ✓, bridge ✓).
4. Click Read URL, Type into the box, Click the counter, or Screenshot — each POSTs a genuine command to 127.0.0.1:17777, which the extension executes on the active tab (the demo page itself).

The page talks to the bridge directly over localhost: Chromium exempts http://127.0.0.1 from mixed-content blocking, and the bridge already sends permissive CORS headers, so an HTTPS GitHub Pages site can reach it. The extension only exposes a presence handshake to pages (so the demo can show "extension installed") — all real actions still go through the token-protected bridge, so a random website can't drive your browser.

Download the extension

• From the site: nexus-ai-bridge-extension.zip
• From GitHub: the source lives in extension/; the packaged zip is at docs/nexus-ai-bridge-extension.zip ("Download raw file").

Load it: Chrome/Edge — unzip, open chrome://extensions, enable Developer mode, "Load unpacked". Kiwi Browser (Android) — ⋮ → Extensions → load the .zip directly.

Pieces

bridge.py              tiny stdlib HTTP server on 127.0.0.1:17777
extension/             Chrome MV3 extension (background + content script)
docs/                  the GitHub Pages site (landing, sandbox, extension zip)
cloud-browser/         hosted-browser alternative you drive from a phone

One-time setup

1. Start the bridge (leave running):

   python C:\Users\EC\ai-browser\bridge.py
   

2. Install the extension:
   • Open chrome://extensions
   • Toggle Developer mode on (top-right)
   • Click Load unpacked
   • Pick C:\Users\EC\ai-browser\extension
3. Connect: click the extension's puzzle-piece icon → pin "AI Browser Bridge" → click its icon → press Connect. Status should turn green ("Connected to bridge").

Calling it from anywhere

Anything that can POST JSON to http://127.0.0.1:17777/command can drive it.

# Click an element by CSS selector
curl -X POST http://127.0.0.1:17777/command \
  -H "Content-Type: application/json" \
  -d '{"action":"click","args":{"selector":"#submit-btn"}}'

# Click by visible text
curl -X POST http://127.0.0.1:17777/command \
  -H "Content-Type: application/json" \
  -d '{"action":"click","args":{"selector":"text=Sign In"}}'

# Type into an input
curl -X POST http://127.0.0.1:17777/command \
  -H "Content-Type: application/json" \
  -d '{"action":"type","args":{"selector":"input[name=q]","text":"hello","submit":true}}'

# Get current page URL/title
curl http://127.0.0.1:17777/command -X POST \
  -H "Content-Type: application/json" \
  -d '{"action":"get_url"}'

Actions

action	args	returns
click	{selector} or {x,y}	element descriptor
type/fill	{selector?, text, append?, submit?}	typed-into element + final value
get_url	—	{url, title}
get_text	{selector?, max_chars?}	{text, length, truncated}
get_html	{selector?, max_chars?}	{html, length, truncated}
query	{selector, limit?}	matching elements descriptors
scroll_to	{selector} or {x,y}	{scrolled_to}
wait_for	{selector, timeout_ms?, state?}	{found}
press	{key, selector?}	{pressed}
screenshot	{format?, quality?}	{data_url} (base64 PNG/JPEG)
navigate	{url}	{url}
list_tabs	—	[{id,title,url,active,windowId}]
switch_tab	{tab_id}	{id, url}

Selector syntax

• CSS selectors: #id, .class, button[type=submit]
• Text match: text=Sign In (exact or substring, deepest match wins)
• Fuzzy text: text~=sign (substring, case-insensitive)

Health / debug

GET  http://127.0.0.1:17777/health
GET  http://127.0.0.1:17777/status

Notes

• The bridge binds to 127.0.0.1 only — not reachable from other machines.
• The extension cannot inject into chrome://, the Chrome Web Store, or other internal pages. Open a normal http(s) tab first.
• Service worker keepalive uses chrome.alarms. The long-poll itself also counts as active work, so the SW stays alive while a poll is in flight.
• Commands without an active tab error out fast — you'll see "no active tab".

Plans

This project is free, and its license stays as is. Support it through GitHub Sponsors:

• Free: use the code, community support.
• Pro ($49/month): priority issues and support.
• Team ($199/month): priority support for your team plus early access.