@shiplightai/mcp2

@shiplightai/mcp2

MCP server for browser UI exploration. Provides tools to launch browser sessions, navigate pages, inspect the DOM, perform actions, and collect locators — designed for coding agents that explore UI then write Playwright tests with @shiplightai/sdk.

Installation

npm install -g @shiplightai/mcp2

Usage

Claude Code

Add to your Claude Code MCP config (~/.claude/settings.json):

{
  "mcpServers": {
    "shiplight": {
      "command": "shiplight-mcp2"
    }
  }
}

Claude Desktop

Add to your Claude Desktop config:

{
  "mcpServers": {
    "shiplight": {
      "command": "npx",
      "args": ["-y", "@shiplightai/mcp2"]
    }
  }
}

Cursor / Windsurf

Add to your MCP config (.cursor/mcp.json or equivalent):

{
  "mcpServers": {
    "shiplight": {
      "command": "npx",
      "args": ["-y", "@shiplightai/mcp2"]
    }
  }
}

Tools

Session Management (5)

Tool	Description
new_session	Launch a browser session with optional device emulation
save_storage_state	Save cookies/localStorage for session reuse
close_session	Close a browser session
close_all	Close all browser sessions
get_session_state	Get current URL and session type

Browser Interaction (7)

Tool	Description
navigate	Navigate to a URL
get_page_info	Get page URL and title (lightweight)
inspect_page	Screenshot + DOM with element indices
act	Perform browser actions using element indices
get_locator	Extract Playwright locator for an element
update_variables	Set session variables for browser tasks
clear_execution_history	Clear action history for a session

Debug (4)

Tool	Description
get_browser_console_logs	Get console logs (errors, warnings)
get_browser_network_logs	Get network request logs
clear_logs	Clear console and network logs
get_artifact	Retrieve screenshots or DOM snapshots

Resources

The server exposes two resources for coding agents:

• shiplight://schemas/action-entity — Parameter docs for all browser actions
• shiplight://docs/agent-sdk — API documentation for @shiplightai/sdk, including a workflow guide for turning UI exploration into test code

Workflow

• Explore — Use inspect_page + act to navigate the app
• Collect locators — Use get_locator on elements you want to test
• Write tests — Use the locators in Playwright tests, with @shiplightai/sdk for AI-powered assertions and self-healing

import { test, expect } from '@playwright/test';
import { createAgent, configureSdk } from '@shiplightai/sdk';

configureSdk({ env: { GOOGLE_API_KEY: process.env.GOOGLE_API_KEY! } });

test('checkout flow', async ({ page }) => {
  const agent = createAgent({ model: 'gemini-2.5-pro' });

  await page.goto('https://myshop.example.com');
  await page.getByRole('button', { name: 'Add to Cart' }).click();
  await agent.assert(page, 'Item was added to cart');
  await agent.run(page, 'Complete checkout with test data');
});

Environment Variables

Variable	Default	Description
WEB_AGENT_MODEL	gemini-2.5-pro	AI model for browser agent
TERMINATION_TIMEOUT	(none)	Session auto-close timeout in ms

License

MIT