Exnos
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Exnosverify the current tab state"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Exnos
Live browser-state verification for AI coding agents.
Your AI says "done." Exnos is how it knows.
One tool call returns the full state of the Chrome tab you're looking at: every field, every button, every console error, network requests, storage, performance—in milliseconds. Read-only. Local. Free to use. Built by GOL Productions.
By GOL Productions.
The Problem
AI coding agents edit files and hope for the best. When something breaks:
You: "The button doesn't work"
AI: "Can you check the console for errors?"
You: "It says TypeError something something"
AI: "Can you paste the full error?"Back and forth. Slow. Frustrating.
Related MCP server: @nimbus21.ai/chrome-devtools-mcp
The Solution
You: "The button doesn't work"
AI: [calls exnos_verify]
AI: "Console shows 'TypeError: handleClick is not defined' at line 47.
The handler was renamed to onClick. Fixing now."Exnos gives your AI eyes. It sees what you see—instantly.
Install
npx @golproductions/exnos@latest setupThat's it. Detects Claude Code, Cursor, and Windsurf—registers with all of them, opens the extension folder, tells you to load it in Chrome. Takes 30 seconds.
1. Connect your agent
{ "mcpServers": { "exnos": { "command": "npx", "args": ["@golproductions/exnos"] } } }Or for Claude Code:
claude mcp add --scope user exnos -- npx @golproductions/exnos2. Load the extension
npx @golproductions/exnos@latest pathOpen chrome://extensions → Developer mode → Load unpacked → select that folder.
Badge reads ON when connected.
What Exnos Can Read
Local pages, always:
localhost,127.0.0.1,*.localhost, and local files.Any other site, only if you allow it: open the site and click the Exnos icon in Chrome's toolbar. Click again to remove it. Your AI cannot allow a site.
Tabs on other sites are not listed, matched or described to your AI.
The console and network capture runs only on sites Exnos can read. A page that was already open when you allowed its site needs one reload before its console and network are captured.
What It Sees
Category | Data |
Identity | URL, title, ready state |
Forms | Every visible field with its live value (passwords, API keys, tokens, card numbers and seed phrases masked) |
Buttons | Text and disabled state |
Checkboxes | Checked state with labels |
Alerts | Visible error/success/warning UI |
Console | Errors (with message and stack), warnings, and failed resource loads since page load, counted separately |
Network | This site's fetch/XHR: URL, status, short response body. Failures first. Other sites' requests only on request. |
WebSocket | Sent and received frames |
Performance | Page load, TTFB, paint timing |
Focus | Which element has focus |
Shadow DOM | Pierces web component boundaries |
Iframes | Same-origin content + cross-origin count |
Screenshot | Optional PNG capture |
Only when asked: localStorage, sessionStorage and cookies (includeStorage, with tokens, keys and session values redacted), requests to other sites (thirdParty), and window.__* app state (appGlobals).
Tools
Tool | Purpose |
| Full live state. The main tool. |
| List the tabs Exnos can read. |
| PNG screenshot of visible tab. |
| Verify multiple tabs at once. |
exnos_verify parameters
Parameter | Description |
| Match tab by URL or title substring. Default: active tab. |
| CSS selector for deep-dive: text, bounds, computed styles, HTML. |
| Include off-screen elements. Default: false. |
| Include requests to other sites. Default: false. |
| Include storage and cookies, credentials redacted. Default: false. |
| Include |
| Also capture PNG of the visible tab. Returns data URL. |
Selector deep-dive
Pass selector to get the following. If nothing matches, the reply is only selectorFound: false.
selectorText— inner text contentselectorVisible— actually visible?selectorBounds—{top, left, width, height}selectorHTML— outer HTMLselectorStyles— computed: color, backgroundColor, fontSize, fontWeight, fontFamily, padding, margin, border, zIndex, overflow, transform, transition
CLI
npx @golproductions/exnos@latest setup # configure everything
npx @golproductions/exnos@latest uninstall # remove everything Exnos added (run it in each project where you ran init)
npx @golproductions/exnos@latest path # extension folder path
npx @golproductions/exnos@latest init # write rules to agent config
npx @golproductions/exnos@latest rules # print rule textArchitecture
┌─────────────┐ MCP/stdio ┌─────────────┐ WebSocket ┌─────────────┐
│ AI Agent │ ◄────────────────► │ MCP Server │ ◄──────────────► │ Extension │
│ │ │ :17872 │ │ │
└─────────────┘ └─────────────┘ └──────┬──────┘
│
Chrome APIs
│
┌──────▼──────┐
│ Your Tab │
└─────────────┘Smart Features
State change detection: Reports
changed: true/falseon repeated callsProxy mode: Multiple agents share one extension connection
Auto-reconnect: Extension reconnects after Chrome/server restart
Failures first: Network errors surface before successful requests
Cross-origin awareness: Reports iframe count even when content isn't readable
Out of the page's reach: what Exnos captures is kept where the page's own scripts cannot read or change it
Untrusted content label: replies tell the AI that page text is data, not instructions
Privacy
Exnos sends nothing to GOL Productions. No account, no telemetry, and no GOL server. Traffic goes from the extension to a local server on 127.0.0.1, which only accepts the Exnos extension and requests from this machine: a web page cannot connect to it.
However: what Exnos returns goes to your AI agent, which forwards it to its model provider. That is why Exnos reads only local pages and the sites you allow, masks passwords and similar fields, and returns storage and cookies only when asked, with credentials redacted. Redaction works by pattern, so it can miss a secret in an unusual place.
Allow the sites you are debugging, not your banking tab.
Notes
Server:
127.0.0.1:17872. The Chrome extension always connects on this port, so keep it free.GET /returns{"exnos":true,"extension":true|false}Console/network capture starts when a page loads on a site Exnos can read; pages already open need one reload
Internal pages (
chrome://) cannot be inspected
License
Free to use, personal or commercial, under the GOL Free License. Exnos is built by GOL Productions: you may use it and share unmodified copies with the credit, but not publish changed versions, build your own product from it, or sell it. What you make with it is yours. See LICENSE.
GOL Productions
Exnos is part of the GOL Productions toolchain.
This server cannot be deployed
Maintenance
Related MCP Connectors
Live browser debugging for AI assistants — DOM, console, network via MCP.
- mcpOAuthcom.screenshotink
Screenshot, diff, audit and sitemap-capture any web page — 5 MCP tools for AI agents.
A paid remote MCP for AI agent browser DevTools MCP, built to return verdicts, receipts, usage logs,
Hosted real Google Chrome MCP with per-user persistent state. Navigate, click, type, screenshot.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceBridges browser content, developer tools data, and web page interactions with Claude through MCP. Enables page inspection, DOM analysis, JavaScript execution, console monitoring, network activity tracking, and screenshot capture across multiple browser tabs.MIT
- AlicenseNot gradedqualityCmaintenanceLets AI coding agents control and inspect a live Chrome browser via MCP, providing Chrome DevTools capabilities for automation, debugging, and performance analysis.9 npmApache 2.0
- AlicenseAqualityDmaintenanceProvides AI coding agents real-time browser access to console logs, network requests, DOM elements, and screenshots via MCP, enabling tight edit-reload-verify feedback loops.155 npm2MIT
- FlicenseNot gradedqualityCmaintenanceEnables AI assistants to control and interact with a Chrome browser via MCP, providing tools for navigation, screenshots, clicking, form filling, content extraction, and tab management.-