GSproxy

GSproxy is a standalone Perl tool that reads a CSV list of remote targets and keeps a working SOCKS proxy alive for each of them. It brings up remote SOCKS servers over an MCP shell bridge, starts local gs-netcat SOCKS clients, watches every proxy, and exposes a single gateway SOCKS port that always points at the first healthy proxy.

Overview

The tool turns a list of endpoint,secret,port rows into a self-healing set of SOCKS proxies. For each row it does three things:

• Brings up a remote SOCKS server through an MCP Tools_Shell_Command call, using a gs-netcat listener wrapped to look like python.
• Starts a local gs-netcat SOCKS client bound to 127.0.0.1 on the row's port.
• Monitors the local SOCKS proxy and logs every status change.

It then listens on a gateway SOCKS port and relays each incoming connection to the first proxy that is currently alive.

The whole tool is written with core Perl modules only, so it runs without extra installs.

Features

• CSV driven config, one proxy per row.
• Remote bring-up through an MCP shell call, reusing the gs-netcat listener command shape.
• Local gs-netcat SOCKS client launched per row.
• Continuous monitoring with configurable interval and per-check timeout.
• Automatic re-bring-up of dead proxies in the background.
• A gateway SOCKS port that follows the first alive proxy.
• Daemon mode with pidfile, status, and stop commands.
• Remote and local teardown that kills leftover gs-netcat processes.
• Offline self-tests that cover config parsing, command building, and the SOCKS handshake.

Requirements

• Perl 5 with the core modules Getopt::Long, IO::Socket::INET, IO::Select, POSIX, HTTP::Tiny, JSON::PP, Time::HiRes, File::Spec, Errno, and File::Basename.
• The gs-netcat binary in PATH, or a path supplied with --gs-netcat.
• Network access from the MCP endpoint so the remote bring-up command can download and run gs-netcat.
• A reachable MCP endpoint that exposes the Tools_Shell_Command tool.

Installation

• Copy gsproxy.pl somewhere on the host and make it executable with chmod +x gsproxy.pl.
• Make sure gs-netcat is installed and on the PATH.
• Create a gsproxy.csv next to the script, or point --config at another file.

Configuration

The config file is CSV. Each non-empty, non-comment line has the form:

endpoint,secret,port[,label]

• endpoint is the base URL of the MCP server that exposes Tools_Shell_Command.
• secret is the gs-netcat shared secret used by both the remote listener and the local client.
• port is the local SOCKS port for that row. It must be a number from 1 to 65535.
• label is an optional friendly name. When missing it defaults to row plus the line number.

Blank lines and lines starting with # are ignored. A header line such as endpoint,secret,port,label is detected and skipped when the third field is not a number. Invalid rows are reported on standard error and skipped.

Example:

endpoint,secret,port,label
https://mcp.example.org,sekret123,20001,alpha
https://mcp.other.org,topsecret9,20002,beta

The endpoint may be a base URL or a full MCP URL. If it does not already end in /gradio_api/mcp or /mcp, the value of --mcp-path is appended.

Usage

Run the tool as a daemon:

./gsproxy.pl

Run one monitor pass and exit:

./gsproxy.pl --once

Run in the foreground:

./gsproxy.pl --no-daemon

Options

The script accepts these command line options:

• --config FILE sets the CSV config path. The default is gsproxy.csv next to the script.
• --gateway-port PORT sets the gateway SOCKS port. The default is 1080.
• --bind ADDR sets the address the gateway listens on. The default is 0.0.0.0.
• --interval SEC sets the monitor interval in seconds. The default is 30.
• --timeout SEC sets the per-check timeout in seconds. The default is 10.
• --mcp-timeout SEC sets the MCP shell command timeout in seconds. The default is 60.
• --test-host HOST sets the host probed through each SOCKS proxy. The default is www.google.com.
• --test-port PORT sets the port probed through each SOCKS proxy. The default is 443.
• --workdir DIR sets the remote working directory for the MCP call. The default is ..
• --gs-netcat PATH sets the gs-netcat binary. The default is gs-netcat from PATH.
• --log FILE sets the log file. The default is gsproxy.log next to the script.
• --pidfile FILE sets the pidfile. The default is gsproxy.pid next to the script.
• --state FILE sets the runtime state file. The default is gsproxy.state next to the script.
• --insecure disables TLS certificate checks for the MCP endpoint.
• --no-mcp skips the remote MCP bring-up call.
• --no-local skips the local gs-netcat client start.
• --no-bringup runs monitor only and never restarts proxies.
• --no-daemon runs in the foreground.
• --once runs one monitor pass and then exits.
• --list parses the config, prints the rows, and exits.
• --status reports whether the daemon is running.
• --stop stops a running daemon and tears down proxies.
• --selftest runs the offline self-tests.
• -h, --help shows the help text.

The --daemon option is a negatable boolean and defaults to on. --verbose and -v increase logging to standard error.

How it works

gsproxy has four moving parts: the remote bring-up, the local client, the monitor, and the gateway.

• Remote bring-up downloads a gs-netcat build with curl, unpacks it, and runs a gs-netcat listener with -lSD -s SECRET. The listening process is renamed to python so it blends into the host.
• The local client runs gs-netcat -D -p PORT -s SECRET, which opens a SOCKS listener on 127.0.0.1 at the row's port.
• The monitor performs a full SOCKS5 handshake against each local port, testing a real CONNECT to a probe host and port. A passing handshake marks the proxy alive.
• The gateway accepts SOCKS5 connections on --bind:--gateway-port and relays them to the first alive proxy listed in the state file.

MCP bring-up

The remote call is an MCP JSON-RPC exchange over HTTP. The script sends an initialize request, captures the Mcp-Session-Id header, sends the notifications/initialized notification, then calls tools/call with the name Tools_Shell_Command. Responses may be plain JSON or server-sent events, so the body is parsed for the last data: line before decoding. Requests are retried up to three times on transport failures.

Monitoring and state

On every interval the monitor checks each proxy, logs only status changes, and writes the list of alive ports to the state file. The first alive port from that list is the current gateway target. When a proxy is found dead, the tool forks a background bring-up unless --no-bringup is set. In --once mode the bring-up runs inline instead of forking.

Gateway relay

The gateway listens, forks per connection, and reads the target port from the state file. It opens a TCP connection to that port and copies bytes in both directions with IO::Select, shutting down each direction as its side closes. If the state file has no valid port, the connection is dropped.

Teardown

Stopping the tool cleans up both sides. Locally it scans /proc for gs-netcat processes that match the -D -p PORT shape and sends them TERM. Remotely it runs a shell command over MCP that scans /proc for processes whose environment contains GSOCKET_ARGS=-lSD -s SECRET and kills them.

Command reference

• Start as a daemon: ./gsproxy.pl
• Start and stop: ./gsproxy.pl --stop
• Check status: ./gsproxy.pl --status
• List config rows: ./gsproxy.pl --list
• One pass, foreground: ./gsproxy.pl --once --no-daemon
• Run self-tests: ./gsproxy.pl --selftest

Files

• gsproxy.csv is the default config file.
• gsproxy.log is the append-only log, one UTC-stamped line per event.
• gsproxy.pid holds the daemon and gateway process ids.
• gsproxy.state holds the first alive port on the first line and all alive ports on the second.

Logging

Every event is written with a UTC timestamp in the form [YYYY-MM-DDTHH:MM:SSZ] message. Bring-up, MCP results, status changes, gateway startup, and teardown are all logged. The log is appended, never truncated, so it grows over time. With --verbose, log lines are also printed to standard error.

Gateway behavior

• The gateway always targets the first alive proxy in listing order.
• It re-reads the state file on every new connection, so failover is immediate.
• A dead or missing target port causes the connection to be dropped without reply.
• Each connection is handled by a forked child, keeping the listener responsive.

Self-tests

The --selftest mode runs offline checks and prints PASS or FAIL per case. It writes a temporary CSV, a temporary state file, and a temporary log under /tmp. It covers config parsing, command string quoting, gs-netcat process discovery, live and dead SOCKS detection, and a full gateway relay against a fake SOCKS server. It exits non-zero if any check fails.

Security notes

• Secrets are masked in logs and list output, so only the first two characters appear.
• The remote process is deliberately renamed to python to hide on the target host.
• The --insecure flag turns off TLS verification and should be used only when the MCP endpoint is trusted.
• The gateway binds to 0.0.0.0 by default. Restrict --bind if the host should not expose the proxy to the network.
• Anyone who can reach the gateway SOCKS port can route through the current proxy, so the port should be firewalled.

Troubleshooting

• If the script reports no usable rows, check that the CSV has valid endpoint,secret,port fields and a numeric port.
• If the daemon will not start, run --status to see whether it is already running, then use --stop first.
• If MCP bring-up fails, confirm the endpoint exposes Tools_Shell_Command and try --insecure only if the certificate is the problem.
• If no proxy becomes alive, run --list to confirm the rows and check that gs-netcat is on the PATH.
• If the gateway has no target, watch the log for SOCKS ... is dead lines to see which proxies are failing.

Credits

Created by Luxferre in 2026, released into the public domain with no warranties.