MCP 服务器

proxyclaw-mcp-py

MCP server giving AI agents controlled residential-proxy web access with optional browser rendering and structured extraction from 60+ sites.

README

ProxyClaw MCP Server (Python)

Model Context Protocol server for ProxyClaw. It gives AI agents controlled residential-proxy web access, optional browser rendering, and structured extraction workflows for legitimate testing, research, and automation use cases.

⚡ Python vs Node.js — Which One?

We ship two MCP servers. Choose based on what you need:

	This repo (Python)	Node.js version
What it does	Browser rendering + structured extraction	Proxy routing + fetch
Best for	JS-rendered pages and structured extraction workflows	Simple fetches, geo-targeting
Networking	Optional TLS-compatible client libraries + Playwright rendering	Standard proxy-backed fetch
Install	`uvx proxyclaw-mcp-server[all]`	`npx proxyclaw-mcp-server`
Tools	6 (+ stealth fetch, render, scrape, extract)	4 (fetch, check_ip, list_countries, rotate)

→ Use Python if you need JavaScript rendering or structured data extraction from supported sites.

→ Use Node.js if you just need to route requests through residential IPs.

Both use the same proxy network — just different levels of power.

Install

# Core proxy workflow
pip install proxyclaw-mcp-server

# Optional compatible HTTP clients
pip install proxyclaw-mcp-server[stealth]

# With browser rendering
pip install proxyclaw-mcp-server[all]

# Or via uv (Claude Desktop recommended)
uv tool install proxyclaw-mcp-server[all]

Claude Desktop Config

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "proxyclaw": {
      "command": "uvx",
      "args": ["proxyclaw-mcp-server[all]"],
      "env": {
        "IPLOOP_API_KEY": "your-api-key"
      }
    }
  }
}

Tools

Tool	What it does
`proxy_fetch_stealth`	Fetch URL through ProxyClaw with optional compatible HTTP clients + retry
`proxy_render`	Headless Chromium render for JavaScript apps and dynamic pages
`proxy_scrape`	Smart cascade: direct → stealth → render → search fallback
`proxy_extract`	Structured data from 60+ sites (eBay, Amazon, LinkedIn, Google, Nasdaq, etc.)
`proxy_check_ip`	Verify exit IP and geo-location
`proxy_list_countries`	List 195+ available countries

Quick Start

Get your API key at iploop.io/signup.

Then ask Claude:

"Fetch https://example.com through a German IP with stealth headers"

"Render https://spa-site.com and extract the product title"

"Scrape eBay search results for 'laptop' and return structured data"

Features

Residential proxy network — 175M+ real IPs across 195+ countries
Compatible HTTP clients — optional curl_cffi / tls_client integrations for more browser-like request behavior
Headless rendering — Playwright rendering for JavaScript-heavy pages
Structured extraction — 60+ site-specific parsers (eBay, Amazon, LinkedIn, Twitter, Nasdaq, etc.)
Smart cascade — tries direct fetch first, escalates to stealth → render → search fallback
Sticky sessions — keep same IP across requests
Cookie persistence — cookies saved between requests
Auto-retry with rotation — configurable retry behavior for transient proxy/network failures

Security & Trust

Source code: https://github.com/Iploop/proxyclaw-mcp-py
Security page: https://proxyclaw.ai/security.html
The package is a thin MCP adapter around the public iploop-sdk; it does not include install hooks, shell execution, credential exfiltration, crypto mining, or obfuscated payloads.
Runtime credential access is limited to IPLOOP_API_KEY, used only to authenticate proxy/API requests.