Skip to main content
Glama

Mobbin Agent

MCP server that gives AI agents browser-based access to Mobbin for design research, screen discovery, and screenshot collection.

Claude Code (AI reasoning + decisions)
    ↕ MCP Protocol (stdio)
Mobbin Agent (TypeScript MCP server)
    ↕ Playwright (browser automation)
Mobbin.com (authenticated session)
    ↕ Local filesystem
data/downloads/ (saved screenshots)

The AI agent decides what to browse, search, and download. The MCP server provides deterministic browser control tools. No extra LLM cost per browser action.

Features

  • 10 MCP tools — lifecycle, navigation, interaction, perception, collection

  • Session persistence — authenticate once, sessions auto-restore across runs

  • Structured extraction — page data returned as classified JSON (screens, flows, apps, categories)

  • Full-resolution downloads — batch download screen images at 1920px PNG quality

  • Anti-detection — realistic user agent, disabled automation flags, human-like delays

  • Direct script access — bypass MCP and import modules directly for batch tasks

Related MCP server: @meshbrow/mcp-server

Requirements

  • Node.js >= 18

  • A Mobbin account (free or paid)

Setup

git clone git@github.com:ismailsaleekh/mobbin-agent.git
cd mobbin-agent
npm install
npx playwright install chromium
npm run build

First-time authentication

Mobbin requires login. On first use, authenticate in a visible browser window — the session is saved automatically for future runs.

node -e "
import('./dist/browser.js').then(async ({ BrowserManager }) => {
  const { MobbinNavigator } = await import('./dist/mobbin.js');
  const browser = new BrowserManager();
  await browser.launch({ headless: false });
  const mobbin = new MobbinNavigator(browser);
  const result = await mobbin.login();
  console.log(result);
  await browser.close();
  process.exit(0);
});
"

A Chromium window opens to mobbin.com/login. Log in manually — the server detects completion and saves cookies to data/session/storage-state.json. Sessions last ~24-48 hours.

Usage with Claude Code

Add to your Claude Code MCP configuration:

{
  "mcpServers": {
    "mobbin": {
      "command": "node",
      "args": ["/absolute/path/to/mobbin-agent/dist/index.js"]
    }
  }
}

Then ask Claude to browse Mobbin:

"Find trending dashboard screen patterns on Mobbin and download the top 20 at full resolution"

"Browse onboarding flows on Mobbin and extract the screen URLs for e-commerce apps"

Tools

10 tools across 5 categories:

Category

Tool

Description

Lifecycle

mobbin_connect

Launch browser and restore saved session

mobbin_login

Navigate to login page for manual authentication

mobbin_disconnect

Save session and close browser

Navigation

mobbin_navigate

Go to any Mobbin URL, wait for load

mobbin_scroll

Scroll page to load more content (infinite scroll)

Interaction

mobbin_click

Click element by text, CSS selector, or coordinates

mobbin_type

Type text into input field

Perception

mobbin_screenshot

Take viewport screenshot (returns PNG)

mobbin_extract

Extract structured data from current page

Collection

mobbin_download

Download screen images by URL to local filesystem

See Tool Reference for parameters, return formats, and examples.

Example: Batch collection script

gather.mjs collects screen images by pattern — bypasses MCP and imports modules directly:

# Collect trending dashboard screens
PATTERN=Dashboard node gather.mjs

# Collect most popular checkout screens with more scrolling
PATTERN=Checkout SORT=mostPopular MAX_SCROLLS=12 node gather.mjs

Images are saved to data/downloads/{pattern}/ at full 1920px PNG resolution.

Project structure

mobbin-agent/
├── src/
│   ├── index.ts          # MCP server entry point (stdio transport)
│   ├── browser.ts        # Playwright browser lifecycle and interactions
│   ├── mobbin.ts          # Mobbin domain logic (login, extract, classify)
│   └── tools.ts           # MCP tool definitions and handlers
├── docs/                  # Full documentation
│   ├── INDEX.md           # Tool reference (all 10 tools)
│   ├── OPERATIONAL-GUIDE.md  # Workflows, rules, recipes
│   ├── connection.md      # MCP setup, session, direct usage
│   ├── tools/             # Detailed tool docs (5 files)
│   ├── reference/         # URL patterns, page types
│   └── troubleshooting/   # Known issues and solutions
├── gather.mjs             # Batch collection example script
├── data/
│   ├── session/           # Saved browser session (gitignored)
│   └── downloads/         # Downloaded images (gitignored)
├── MOBBIN-INSTRUCTIONS.md # Mandatory reading for AI agents
├── package.json
└── tsconfig.json

Documentation

Document

Purpose

MOBBIN-INSTRUCTIONS.md

Entry point — key rules for AI agents

Tool Reference

All 10 tools with links to detailed docs

Operational Guide

Rules, collection workflows, script templates, timeouts

Connection Guide

MCP config, browser lifecycle, session persistence, direct script usage

URL Patterns

Search templates, filter catalogs, CDN URLs

Page Types

Page classification reference

Troubleshooting

Known issues and solutions

Tech stack

Component

Technology

MCP server

@modelcontextprotocol/sdk 1.12

Browser automation

playwright 1.52 (Chromium)

Language

TypeScript 5.8 (ESM, strict)

Transport

stdio (JSON-RPC 2.0)

Session storage

Playwright storageState (JSON cookies)

License

MIT

Available Tools

10 tools
mobbin_clickA

Click on an element in the current page. Provide one of: text (click element containing text), selector (CSS selector), or x+y coordinates.

Inputs: x, y, text, selector.

mobbin_connectA

Launch a browser and connect to Mobbin. Restores saved session if available. Use headless=false for first-time login (so you can see the browser window). After connecting, use mobbin_login if not authenticated.

Inputs: headless.

mobbin_disconnectA

Close the browser and save the current session. The session can be restored on the next mobbin_connect call.

Inputs: no named parameters.

mobbin_downloadA

Download one or more images by URL to the local filesystem. Images are saved to the downloads directory: mobbin-agent/data/downloads/ Returns the list of saved file paths.

Inputs: urls (required), prefix.

mobbin_extractA

Extract structured data from the current Mobbin page. Returns: page type, headings, all links (with Mobbin-specific filtering for screens/flows/apps/categories), all images (with design-relevant image filtering), and a text preview. Use this to understand page content without taking a screenshot.

Inputs: no named parameters.

mobbin_loginA

Navigate to Mobbin login page and wait for the user to authenticate manually. The browser must be launched in non-headless mode (mobbin_connect with headless=false). Waits up to 5 minutes for login completion, then saves session cookies.

Inputs: no named parameters.

mobbin_navigateA

Navigate to a Mobbin URL and wait for the page to load. Returns page title and final URL. Known URL patterns: {"home":"https://mobbin.com","login":"https://mobbin.com/login","exploreWeb":"https://mobbin.com/explore/web","exploreMobile":"https://mobbin.com/explore/mobile","browseWebScreens":"https://mobbin.com/browse/web/screens","browseMobileScreens":"https://mobbin.com/browse/mobile/screens","browseWebFlows":"https://mobbin.com/browse/web/flows","browseMobileFlows":"https://mobbin.com/browse/mobile/flows"} Screen categories: home, dashboard, login, signup, onboarding, checkout, product-detail, browse-discover, search, settings-preferences, profile, chat-detail, notifications, loading, success, error, empty-state, pricing, payments, orders Flow categories: logging-in, signing-up, onboarding, adding-to-cart-bag, editing-profile, searching, filtering-sorting, subscribing-upgrading, chatting-sending-messages, sharing, uploading, booking Screen category URL: /explore/{mobile|web}/screens/{category} Flow category URL: /explore/{mobile|web}/flows/{category} Screen detail URL: /explore/screens/{UUID}

Inputs: url (required).

mobbin_screenshotA

Take a screenshot of the current browser viewport. Returns a PNG image. Use this to see what is currently displayed on the Mobbin page.

Inputs: fullPage.

mobbin_scrollA

Scroll the current page up or down. Useful for loading more content (Mobbin uses infinite scroll). Returns scroll position.

Inputs: amount, direction.

mobbin_typeA

Type text into an input field. Optionally specify a CSS selector for the target input. Can press Enter after typing.

Inputs: text (required), clear, selector, pressEnter.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.4/5.0

    Scored across 10 tools

    Disambiguation5/5

    Each tool has a distinct responsibility: session lifecycle (connect/login/disconnect), navigation, observation (screenshot/extract), interaction (click/type/scroll), and asset retrieval (download). There is no meaningful overlap between tools, and any apparent boundaries like connect vs navigate are clearly explained.

    Naming Consistency5/5

    All tool names follow the same mobbin_<verb> pattern using lowercase snake_case. The verbs are direct and describe exactly what each tool does, making the set predictable and easy for an agent to reason about.

    Tool Count5/5

    Ten tools is a well-scoped set for a browser automation server focused on Mobbin. Each tool contributes a necessary capability without redundancy, and the count feels appropriate for the workflow of connecting, navigating, extracting, and downloading.

    Completeness5/5

    The tool surface covers the full lifecycle needed to work with Mobbin: session management, manual authentication, navigation to known screens/flows, page inspection, interaction, scrolling for infinite content, screenshots, and downloading images. There are no critical missing operations for the apparent domain.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      MCP server for Meshbrow that gives AI agents full browser automation capabilities with stealth anti-detection, enabling natural language control of browser sessions, data extraction, and multi-browser fleets.
      MIT
    • F
      license
      B
      quality
      D
      maintenance
      An MCP server that enables AI coding tools to control a browser for automated actions, UI extraction, network interception, and screenshots.
      23
      1
      -