Skip to main content
Glama
fusionproxy-mcp

fusionproxy-mcp

Official

Fusion Proxy MCP server: web access for AI agents through residential proxies

An MCP server that lets Claude, Cursor, VS Code Copilot and any other MCP client fetch web pages through residential IPs in the country you choose. Useful when an agent gets blocked, rate-limited, or needs to see a page the way a visitor from another country sees it.

Tools

Tool

What it does

fetch_page(url, country?, session?, output?, max_chars?)

Downloads a public page through a residential IP and returns it as Markdown (default), plain text or HTML

check_exit_ip(country?, session?)

Shows the IP, city, country and network that websites see

  • country: two-letter code (us, de, jp, ...). Leave it out for any country.

  • session: any name. Calls with the same name keep the same IP for 10 minutes, so an agent can browse several pages of one site as one visitor.

Example conversation:

You: What does the pricing page of example.com show to visitors from Japan?

The agent calls fetch_page(url="https://example.com/pricing", country="jp") and reads the Markdown.

What the calls return (IP addresses and networks are placeholders; yours will differ):

check_exit_ip {'country': 'jp'}                  -> IP 198.51.100.7, Tokyo, Tokyo, JP (AS64500 Example ISP)
check_exit_ip {'country': 'us', 'session': 'a'}  -> IP 203.0.113.42, Austin, Texas, US (AS64501 Example Broadband)
check_exit_ip {'country': 'us', 'session': 'a'}  -> IP 203.0.113.42, Austin, Texas, US (AS64501 Example Broadband)
fetch_page(https://quotes.toscrape.com/, country=gb) -> Status: 200 ... # [Quotes to Scrape](/) ...

The same session gives the same IP; a new session or no session gives a new one.

Related MCP server: GeoRanker Web Scraping MCP

Setup

You need Fusion Proxy credentials (username and password from the dashboard; there is a free trial) and uv (or pip).

Claude Code

claude mcp add fusionproxy -e FUSION_USER=your_username -e FUSION_PASS=your_password \
  -- uvx --from git+https://github.com/FusionProxies/fusionproxy-mcp fusionproxy-mcp

Claude Desktop

claude_desktop_config.json (Settings, Developer, Edit Config):

{
  "mcpServers": {
    "fusionproxy": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/FusionProxies/fusionproxy-mcp", "fusionproxy-mcp"],
      "env": { "FUSION_USER": "your_username", "FUSION_PASS": "your_password" }
    }
  }
}

Cursor

~/.cursor/mcp.json: the same mcpServers block as for Claude Desktop.

VS Code

.vscode/mcp.json:

{
  "servers": {
    "fusionproxy": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--from", "git+https://github.com/FusionProxies/fusionproxy-mcp", "fusionproxy-mcp"],
      "env": { "FUSION_USER": "your_username", "FUSION_PASS": "your_password" }
    }
  }
}

Without uv

pip install git+https://github.com/FusionProxies/fusionproxy-mcp
fusionproxy-mcp   # speaks MCP over stdio; point your client's "command" at it

Settings

Variable

Required

Meaning

FUSION_USER

yes

Proxy username (main user or a sub-user)

FUSION_PASS

yes

Proxy password

FUSION_COUNTRY

no

Default country when a call does not pass one

FUSION_GATEWAY

no

Gateway host:port, default resi.fusionproxy.net:12286

Traffic through the server counts against your residential plan like any other request. A typical HTML page is 50 KB to 2 MB.

How it works

Each call opens a connection to the Fusion Proxy gateway with the options in the username (-country-jp, -sessid-...-sessTime-600s), downloads the page with a normal browser User-Agent, follows redirects and converts HTML to Markdown with html2text. If a residential connection drops, the call retries once on a new connection. Only public http and https URLs are accepted: localhost, .local names and private, loopback or reserved IP addresses are refused.

The server does not run JavaScript. For pages that render in the browser, see playwright-proxy.

Works with the MCP Python SDK 1.x and 2.x.

Development

pip install -e ".[test]"
pytest            # no network and no credentials needed

Changes are listed in CHANGELOG.md.

Use responsibly

Fetch public pages only, respect the sites' terms and robots.txt, and keep the request rate reasonable.

License

MIT, see LICENSE.

Available Tools

2 tools
check_exit_ipB

Show the IP address, location and network that websites see through the proxy.

Args: country: two-letter country code of the exit IP. Optional. session: session name to check a sticky IP. Optional.

Inputs: country, session. Returns: result

fetch_pageA

Fetch a public web page through a residential IP and return its content.

Args: url: http(s) URL of the page. country: two-letter country code of the exit IP (for example "us", "de", "jp"). Optional. session: any name; calls with the same name keep the same IP for 10 minutes. Optional. output: "markdown" (default, readable), "text" or "html" (raw). max_chars: cut the result to this many characters (default 20000).

Inputs: url (required), output, country, session, +1 more. Returns: result

Tool Schema Changelog

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

Loading tool schema changelog…

    TDQS

    A3.8/5.0

    Scored across 2 tools

    Disambiguation5/5

    fetch_page retrieves page content while check_exit_ip is a pure diagnostic that reports the exit IP/geo; their purposes are unmistakably distinct despite sharing the optional country/session parameters. An agent would never confuse the two.

    Naming Consistency5/5

    Both tools follow the same snake_case verb_noun convention (fetch_page, check_exit_ip), with verbs that accurately describe the action. No mixing of styles or vague names.

    Tool Count3/5

    Two tools is on the thin side; fetch plus a single diagnostic leaves little room for related operations. The scope is coherent, but the surface is closer to a minimal wrapper than a well-rounded toolset.

    Completeness4/5

    The core operation (proxied page fetch) and the key verification step (exit IP check) are both present, covering the main workflow. Minor gaps exist: no non-GET requests, no session management (list/clear/expire), and the sticky-IP duration is fixed.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI agents to fetch and extract clean, readable content from web pages, and search within pages for specific queries, without needing a full browser.
      1
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables AI assistants to fetch public web pages as readable text or HTML, with caching, live refresh, and parallel job handling.
      3
      17 npm
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Enables AI agents to read any web page as clean Markdown, pull title/description/preview metadata, capture desktop, mobile, or full-page screenshots, and list every link or image on a URL. JavaScript-heavy pages are rendered in a real browser and blocked requests retried through a proxy, so agents can reach content they'd otherwise miss.
      5
      61 npm
      MIT
    • A
      license
      A
      quality
      C
      maintenance
      Enables AI agents to fetch any web page and receive clean, token-efficient Markdown, escalating only as needed from direct HTTP with browser TLS fingerprints to headless-browser rendering and residential proxies when a site blocks access. Supports question-focused passage retrieval with BM25 ranking, pagination, caching, JS rendering control, and safe-by-default restrictions on local and private-network URLs.
      1
      AGPL 3.0