nodemaven-mcp

nodemaven-mcp

MCP server that lets AI agents route web requests through NodeMaven residential and mobile proxies, with tools for geo-targeted fetching, proxy checking, and location/usage queries.

Category
访问服务器

README

<img src="docs/img/banner.png" alt="nodemaven-mcp" width="100%">

nodemaven-mcp

CI Python 3.10+ License: MIT

English | Русский

Give your AI agent a residential IP. An MCP server that lets Claude Code, Claude Desktop, Cursor or any other MCP client route web requests through NodeMaven residential and mobile proxies: pick an exit country, verify what the target site sees, and pull the page.

you:    read this product page the way a shopper in Munich sees it
agent:  nodemaven_fetch(url=..., country="de", city="munich")
        -> 200, de, 14.99 EUR, exit IP 91.34.x.x (Vodafone), 812 ms

Unofficial, community-built, MIT licensed. Not affiliated with NodeMaven.


Why

Agents are increasingly the thing doing the browsing, and they hit the same wall everyone else does: datacenter IPs get blocked, and prices, stock and search results change by geography. The proxy is already the fix - it just isn't reachable from inside the agent's session.

This server closes that gap in five tools. Nothing here needs a scraping DSL: the agent asks for a country, gets a working connection, and reads the page.

Tools

Tool What it does Needs
nodemaven_build_proxy_url Builds a proxy URL with geo targeting encoded in the username, ready for curl, Playwright, Scrapy or requests proxy creds
nodemaven_check_proxy One request through the proxy: exit IP, geo, ISP, latency proxy creds
nodemaven_fetch Fetches a URL through the proxy, HTML converted to readable text proxy creds
nodemaven_list_locations Countries, regions, cities, ISPs, zip codes available for targeting API key
nodemaven_get_usage Traffic consumed, request counts, per-domain breakdown API key

Every tool takes response_format: markdown (default, compact) or json (complete, for programmatic use), and paginates with limit / offset.

Install

git clone https://github.com/Artirain/nodemaven-mcp
cd nodemaven-mcp
pip install -e .

Requires Python 3.10+.

Configure

Two independent sets of credentials, both from the NodeMaven dashboard:

  • NODEMAVEN_API_KEY - Profile -> API Key. Used by the location and usage tools.
  • NODEMAVEN_PROXY_USERNAME / NODEMAVEN_PROXY_PASSWORD - Proxy Setup. Used by everything that sends traffic through the proxy.

You can set only one set; tools that need the other say so instead of failing silently.

Claude Code

claude mcp add nodemaven \
  -e NODEMAVEN_API_KEY=... \
  -e NODEMAVEN_PROXY_USERNAME=... \
  -e NODEMAVEN_PROXY_PASSWORD=... \
  -- python -m nodemaven_mcp

Claude Desktop / Cursor

claude_desktop_config.json (or .cursor/mcp.json):

{
  "mcpServers": {
    "nodemaven": {
      "command": "python",
      "args": ["-m", "nodemaven_mcp"],
      "env": {
        "NODEMAVEN_API_KEY": "your-api-key",
        "NODEMAVEN_PROXY_USERNAME": "your-proxy-username",
        "NODEMAVEN_PROXY_PASSWORD": "your-proxy-password"
      }
    }
  }
}

A local .env in the working directory also works - see .env.example. Environment variables always win over the file.

On Windows, give the absolute path to python.exe rather than bare python. If Python came from the Microsoft Store, the python on your PATH is an app execution alias that MCP clients cannot spawn, and the server silently never appears. python -c "import sys; print(sys.executable)" prints the path to use.

Use it

Three things worth asking your agent, once it is connected:

Check what a site sees

"Am I actually coming out of Germany? Use a Vodafone IP if you can."

The agent calls nodemaven_check_proxy(country="de", isp="vodafone") and reports the exit IP, city and latency. Useful as a smoke test before a long run.

Read geo-restricted content

"Fetch this page as a US shopper in New York and tell me the price."

nodemaven_fetch(url=..., country="us", city="new york") returns readable text, truncated to 200 KB by default so it does not flood the context.

Keep one identity across a flow

"Log into the demo account and walk through checkout, same IP the whole time."

Pass the same session_id (4-10 alphanumerics) on every call and NodeMaven pins the exit IP; drop it and the IP rotates per request.

Targeting spellings matter. nodemaven_list_locations(level="cities", country="de") gives the exact values NodeMaven accepts, which is the difference between a working pool and an empty one.

See what it actually does

A tool call is easy to trust and hard to verify. examples/traffic_dashboard.py runs a local forward proxy with a live dashboard, so you can point the server at it and watch every hop it makes:

python examples/traffic_dashboard.py     # proxy on :8080, dashboard on :8099
NODEMAVEN_PROXY_HOST=127.0.0.1
NODEMAVEN_HTTP_PORT=8080

Dashboard showing proxy hops with decoded targeting

Each row is a real network request. The route column is decoded from the proxy username, which is exactly the string NodeMaven uses to choose the exit IP - so you can confirm that country="de", city="Munich" in a tool call really did leave your machine as DE / MUNICH, and that a session_id stayed pinned across calls.

The dashboard is a development aid: it accepts any credentials and does not encrypt anything. Do not put it on a network you share.

Design notes

Passwords stay masked. build_proxy_url returns *** in place of the password unless the caller passes reveal_credentials=true. Transcripts get shared, pasted into issues and fed back into models; a proxy password should not ride along by default.

Errors are written for an agent, not a log file. A refused connection says that an over-narrow city or ISP filter is the usual cause and suggests widening it. A 401 names the environment variable to check. An agent that can read the error can retry correctly on the first attempt.

Targeting is pure and tested. NodeMaven encodes location in the proxy username:

acct-country-us-region-california-city-los_angeles-isp-t_mobile_usa-sid-ab12cd-filter-medium

Building that string - normalizing "Los Angeles" to los_angeles, rejecting a session id that would silently disable stickiness, validating the port against the range for the chosen protocol - is pure logic, so it is covered by tests that never touch the network.

No parser dependency. HTML-to-text runs on html.parser from the standard library. The dependency list is mcp, httpx and pydantic.

Development

pip install -e ".[dev]"
pytest -q          # 60 tests, no network, no credentials needed
ruff check .

CI runs the suite on Python 3.10, 3.11 and 3.12.

Contributions welcome - see CONTRIBUTING.md.

License

MIT

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选