eyebrowse
Eyebrowse is an MCP server that gives AI agents eyes and hands in your real Chrome browser, allowing them to see screenshots, read page structure, and interact with pages like clicking, typing, scrolling, and navigating.
README
Eyebrowse
Give your AI agent eyes and hands in the browser.
Eyebrowse is an MCP server that connects AI coding agents to real Chrome browsers. Not headless. Not isolated. Your browser, your session, AI-assisted.
What is Eyebrowse?
Eyebrowse bridges AI agents and Chrome through the Model Context Protocol (MCP). Your AI sees screenshots, reads page structure via accessibility trees, and interacts with pages -- clicking, typing, scrolling, navigating -- all in a real Chrome window you can watch and use alongside it.
Unlike headless browser tools, Eyebrowse operates in your actual browser. Same cookies, same logins, same tabs. The AI is a copilot in your browser, not a robot in a hidden one.
Why Eyebrowse?
| Eyebrowse | Playwright/Puppeteer MCP | Headless agents | |
|---|---|---|---|
| Browser | Your real Chrome | Hidden headless browser | Hidden headless browser |
| Session | Shared -- your cookies, logins, state | Isolated -- starts fresh | Isolated |
| Visibility | You see everything the AI does | Invisible | Invisible |
| Interaction | AI and human use the same window | AI only | AI only |
| Setup | Extension + daemon | npm package | Varies |
Use Cases
- Web scraping -- "Search for X on these 5 sites, extract prices and links" (supports parallel workers with separate windows)
- QA review -- "Look at this page and tell me what's broken"
- Form filling -- "Fill out this application with these details"
- Research -- "Find recent articles about Y, summarize the top 3"
- Page capture -- "Save this page as a PDF" or "Take a screenshot of the dashboard"
- Debugging -- "Open localhost:3000, click the login button, tell me what happens"
- Monitoring -- watch a page for changes using your existing logged-in session
- Workflow automation -- automate repetitive browser tasks in your authenticated session
Features
- Screenshots -- capture visible tab or full-page screenshots as JPEG
- Accessibility tree -- read page structure via Chrome DevTools Protocol
- Click, type, scroll, navigate -- full browser interaction by CSS selector, text content, or coordinates
- Multi-window and multi-tab -- pair specific windows to sessions, list and switch tabs
- Highlights and annotations -- overlay elements with colored highlights, callout annotations, or toast messages
- Print to PDF -- export pages with configurable paper size, orientation, and page ranges
- Evaluate JavaScript -- run expressions in the page context and return structured data, ideal for bulk DOM extraction
- Auto-start daemon -- MCP server spawns the daemon automatically on first tool call
- Session isolation -- each Claude Code session registers independently with its own pairing
Quick Install
npx eyebrowse install
The installer checks prerequisites, configures Claude Code's MCP settings, and walks you through loading the Chrome extension. See Manual Setup if you prefer to do it yourself.
Manual Setup
1. Install Bun
curl -fsSL https://bun.sh/install | bash
2. Clone the repository
git clone https://github.com/evan108108/eyebrowse.git
cd eyebrowse
3. Install dependencies
bun install
4. Load the Chrome extension
- Open
chrome://extensionsin Chrome - Enable Developer mode (top right)
- Click Load unpacked
- Select the
extension/directory from this repository
5. Add the MCP server to Claude Code
Add the following to ~/.claude.json (or your project's .mcp.json):
{
"mcpServers": {
"eyebrowse": {
"command": "bun",
"args": ["run", "/path/to/eyebrowse/mcp/index.ts"]
}
}
}
Replace /path/to/eyebrowse with the actual path where you cloned the repo.
Architecture
+-------------------+ WebSocket +-------------------+ stdio/MCP +-------------------+
| | ws://localhost:7890/ws | | | |
| Chrome Extension |<---------------------->| Daemon (Bun) |<--------------------->| MCP Server |
| (Manifest V3) | | Port 7890 | HTTP JSON API | (stdio transport)|
| | events, bridge req/res| | | |
+-------------------+ +-------------------+ +-------------------+
| | |
v v v
chrome.tabs API Session registry Claude Code
chrome.debugger API Window pairing (MCP client)
chrome.scripting API Auth token management
Content script injection Rate limiting
Chrome Extension -- Manifest V3 service worker that connects to the daemon over WebSocket. Bridges Chrome APIs (tabs, debugger, scripting) to daemon requests. Handles screenshots, accessibility tree capture, page interactions, overlays, and PDF export. Maintains a persistent connection with keepalive pings and exponential backoff reconnection.
Daemon -- Bun HTTP + WebSocket server on port 7890. Routes tool calls from the MCP server to the extension, manages session registration, window pairing, and authentication. Runs a stale session reaper every 2 minutes.
MCP Server -- Thin MCP-protocol wrapper that exposes 21 tools over stdio. Communicates with the daemon via HTTP. Auto-starts the daemon if unreachable. Each instance registers as a session identified by the working directory name.
Available Tools
| Tool | Description |
|---|---|
look |
Screenshot + page URL/title. Optionally include the accessibility tree. |
screenshot |
Capture a screenshot of the current page. Supports full-page and quality settings. |
get_page_info |
Get current page URL and title. Optionally include the accessibility tree. |
browser_status |
Check connection health and paired window info. |
list_windows |
List all Chrome windows. |
pair_window |
Pair with a Chrome window by its window ID. |
unpair |
Release the currently paired Chrome window. |
open_browser |
Open a new Chrome window at the given URL. Auto-pairs to the session. |
click |
Click an element by CSS selector, visible text, or x/y coordinates. |
type |
Type text into an element. Supports clearing, pressing Enter, and keystroke delay. |
scroll |
Scroll the page by direction/amount, to an edge, or scroll an element into view. |
navigate |
Navigate to a URL, or go back/forward/reload. |
highlight |
Highlight elements matching a CSS selector with a configurable color. |
annotate |
Add a callout annotation near an element. |
show_message |
Show a toast notification in the browser. |
clear_overlays |
Remove all highlights and annotations from the page. |
list_tabs |
List all tabs in the paired Chrome window. |
switch_tab |
Switch to a specific tab by its tab ID. |
close_window |
Close the paired window or a specific window by ID. |
evaluate |
Execute a JavaScript expression in the page and return the result. Use for bulk DOM data extraction. |
print_to_pdf |
Export the current page as a PDF file. Returns the file path. |
How It Works
-
Connect -- The Chrome extension opens a WebSocket connection to the daemon at
ws://localhost:7890/ws. The daemon validates the origin is achrome-extension://URL. -
Register -- When Claude Code starts a session, the MCP server registers with the daemon over HTTP, receiving a session ID and bearer token. If the daemon is not running, the MCP server spawns it automatically.
-
Pair -- A Chrome window is paired to the session, either from the extension popup or programmatically via
open_browserorpair_window. Each window can only be paired to one session at a time. -
Execute -- Claude Code calls MCP tools. The MCP server forwards them as HTTP requests to the daemon, which translates them into bridge requests over WebSocket to the extension. The extension executes them using Chrome APIs and returns results back through the same chain.
Security
Eyebrowse operates entirely locally. Your data never leaves your machine. No cloud services, no telemetry, no external connections.
- Local-only -- The daemon binds to
localhost:7890. The extension connects only tows://localhost:7890/ws. - Token authentication -- The daemon generates a random UUID token on first run, stored at
~/.eyebrowse/daemon.tokenwith0600permissions. All HTTP requests require this token as a Bearer header. Constant-time comparison prevents timing attacks. - Origin validation -- WebSocket upgrades are rejected unless the origin starts with
chrome-extension://. - CDP allowlist -- Only four Chrome DevTools Protocol commands are permitted:
Accessibility.getFullAXTree,Accessibility.getPartialAXTree,Page.printToPDF, andRuntime.evaluate. - URL validation --
open_browserandnavigateonly accepthttp:andhttps:URLs. - Rate limiting --
open_browseris limited to 3 calls per minute per session. - Session isolation -- Each MCP server instance registers as a separate session with independent window pairing.
Development
Run the daemon
bun run daemon/index.ts
The daemon starts on http://localhost:7890. Logs go to stderr. A daemon token is created at ~/.eyebrowse/daemon.token on first run.
Run the MCP server (standalone)
bun run mcp/index.ts
The MCP server communicates over stdio. It will auto-start the daemon if it is not already running.
Optional: launchd (macOS auto-start)
A template plist is provided at com.eyebrowse.daemon.plist. Replace the placeholder paths (__BUN_PATH__, __EYEBROWSE_HOME__, __BUN_DIR__) and copy to ~/Library/LaunchAgents/ to have the daemon start at login.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.