Skip to main content
Last Updated: January 8, 2026
Deploy from the SaladCloud Portal.

Overview

MCProxy enables AI agents to control browsers running on geographically distributed SaladCloud containers. The browser server runs on SaladCloud, while the MCP server runs locally on your machine and connects to AI agent clients like Claude Desktop, Claude Code, Cursor, or Windsurf. Use cases include:
  • Regional price checking - Compare prices from different geographic locations
  • Geo-targeted content verification - Test how content appears in different regions
  • Distributed web automation - Automate browser tasks across multiple locations
  • Multi-browser testing - Test websites across Chromium, Firefox, and WebKit (Safari)

Key Features

  • Multi-browser support: Chromium (default), Firefox, or WebKit (Safari)
  • Mobile device emulation: 100+ device profiles (iPhone, iPad, Pixel, Galaxy, etc.)
  • Stealth mode: WebGL spoofing, navigator overrides, and realistic browser fingerprints
  • Humanized interactions: Human-like clicks, typing, and scrolling to avoid bot detection
  • Cloudflare auto-wait: Automatically wait for challenges to complete
  • Cookie persistence: Save and restore cookies for session management
  • Location-aware sessions: Each session reports geographic location (city, country, timezone, ISP)
  • 33 MCP tools: Full browser automation including coordinate-based clicking for vision agents

Connecting Your MCP Client

After deploying the recipe, configure your MCP client to connect using the Docker-based MCP server.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

Claude Code

Add to your workspace .mcp.json or global settings:

Cursor / Windsurf

Add to your MCP configuration file:
Replace your-auth-token with the Auth Token you configured when deploying the recipe, and your-endpoint.salad.cloud with your container group’s DNS endpoint.

Example Usage

Once your MCP client is connected, you can ask your AI agent to perform browser automation tasks:

Basic Session Creation

Mobile Device Emulation

Multi-Browser Testing

Geo-Distributed Price Comparison

Humanized Browsing

Cloudflare-Protected Sites

MCP Tools Reference

Session Management

Interaction (Selector-Based)

Interaction (Coordinate-Based)

Uses relative coordinates (0-1 range) for vision agents:

Content Extraction

How To Use This Recipe

Authentication

The Auth Token you provide when deploying the recipe must match the MCPROXY_AUTH_TOKEN configured in your MCP client. This shared secret authenticates the connection between your local MCP server and the browser servers running on SaladCloud.

Replica Count

The recipe is configured for 3 replicas by default. Each replica can handle multiple browser contexts (10 by default). More replicas give you access to more geographic locations for distributed browsing.

Configuration Options

Deploy And Connect

  1. Deploy the MCProxy recipe from the SaladCloud Portal
  2. Enter a container group name and your Auth Token
  3. Wait for instances to enter the running state
  4. Note your container group’s DNS endpoint (e.g., vegetable-words-abc123.salad.cloud)
  5. Configure your MCP client using the endpoint with wss:// prefix
  6. Start creating browser sessions through your AI agent

Source Code

GitHub Repository Recipe Source