viyv-browser

viyv-browser

English | 日本語

Chrome Extension + MCP Server for AI agent browser automation with semantic automation capabilities.

Enables AI agents to operate in the user's real browser with authentic login sessions — no headless browsers, no credential sharing.

Why viyv-browser?

Standard browser tools (like Claude in Chrome) let AI agents interact with web pages. viyv-browser goes further — it adds a semantic automation layer that makes browser automation reliable, repeatable, and scalable.

	Claude in Chrome	viyv-browser
Data extraction	AI analyzes the full page each time — can be inaccurate on complex pages	Define targets to extract exactly the data you need, every time
Reproducibility	AI decides how to act each time — results may vary	Save semantic definitions and replay the same operation reliably
Batch execution	Manual, one request at a time	Jobs run multiple scenarios in sequence with error handling
Scheduled execution	Not available	Run Jobs on a schedule via Cowork
History & comparison	None	Execution history saved, diffs between runs auto-detected
Dashboard	None	Chart.js-powered interactive visualizations, auto-generated
Export	None	CSV / JSON report export
Google Sheets	Not available	Read, write, and navigate Sheets with session cookies (no OAuth)
Low-cost page interaction	Screenshot-based (high token cost)	Indexed DOM text — no screenshots needed, works with text-only LLMs

In short: Use standard browser tools for one-off tasks. Use viyv-browser when you need repeatable execution, accurate data collection, batch processing, scheduled monitoring, or Google Sheets integration.

Use Cases

Recurring data collection & price monitoring — Collect product prices from e-commerce sites daily. Target definitions extract prices accurately, Jobs batch across multiple sites, history comparison detects price changes.

Competitive analysis — Collect competitor product info and reviews on a schedule, visualize trends in a Dashboard, export to spreadsheets via CSV or write directly to Google Sheets.

E2E test automation — Define login → action → verification as a Scenario. Run Jobs from CI/CD pipelines and save test results as reports.

Web scraping — Target-based extraction pulls exactly the data you need. Loop support handles pagination. Authenticated sessions access protected content.

Business process automation — Turn repetitive tasks (data entry, form submission, report retrieval) into Scenarios and run them automatically.

Google Sheets integration — Read, write, and navigate Google Sheets using browser session cookies — no OAuth setup required. Extract data from websites and write directly to spreadsheets, or read spreadsheet data for use in automation workflows.

Quick Start

Step 1: Install the Chrome Extension

1. Download viyv-browser-extension.zip from the latest release
2. Unzip the file
3. Open chrome://extensions/ in Chrome
4. Enable Developer mode (top-right toggle)
5. Click Load unpacked and select the unzipped folder

Step 2: Setup

Choose your AI client:

Claude Desktop / Cowork

npx -y viyv-browser-mcp@latest setup --claude-desktop

Restart Claude Desktop after running setup.

Plugin: Add marketplace from GitHub (BrainFiber/viyv-browser) and install viyv-browser.

Claude Code

npx -y viyv-browser-mcp@latest setup

Plugin: Add marketplace from GitHub (BrainFiber/viyv-browser) and install viyv-browser-code. The plugin includes .mcp.json so the MCP server starts automatically — no claude mcp add needed.

Other MCP Clients (Cursor, etc.)

First, run setup to register the Native Messaging Host:

npx -y viyv-browser-mcp@latest setup

Then add to your client's MCP configuration:

{
  "mcpServers": {
    "viyv-browser": {
      "command": "npx",
      "args": ["-y", "viyv-browser-mcp@latest"]
    }
  }
}

Step 3: Verify the Connection

1. Start your AI client (this launches the MCP Server automatically)
2. The Chrome extension popup should show Connected
3. Ask your AI agent: "Check the browser connection status"

Semantic Automation

viyv-browser's semantic automation system lets you define what to automate, not how. Definitions are saved and reused across sessions.

Architecture

Page (URL pattern identifies the site)
  └─ Target (locate buttons, inputs, text elements)
       ├─ Action (click, type, select — step-by-step recipes)
       └─ Fetch (data extraction rules)
            └─ Scenario (multi-page workflow combining actions & fetches)
                 └─ Job (batch execution of multiple scenarios)

Page & Target