TCP Socket MCP Server

A Model Context Protocol (MCP) server that provides raw TCP socket access, enabling AI models to interact directly with network services using raw TCP Sockets. Supports multiple concurrent connections, buffering of response data and triggering automatic responses.

Motivation and Context

Many network services and IoT devices communicate via raw TCP protocols that aren't covered by existing HTTP-based MCP servers. TcpSocketMCP enables:

• Direct interaction with embedded devices and IoT systems
• Network protocol debugging and testing
• Legacy system integration without HTTP wrappers
• Protocol reverse engineering and analysis
• Automated responses via trigger patterns (useful for IRC, telnet, custom protocols)

This addresses the need for low-level network access that several community members have expressed, particularly for industrial automation, IoT development, and network security testing scenarios.

Demo

Interrogating a device to figure out what it is

Sending data to the device

Sample output from TCP interactions

Installation & Setup

Install from PyPI

# Install with pip
pip install TcpSocketMCP

# Install with uv (recommended)
uv add TcpSocketMCP

# Add to Claude Code (recommended)
claude mcp add rawtcp -- uvx TcpSocketMCP

For Claude Desktop

Add the server to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

Option 1: Using installed package (recommended)

{
  "mcpServers": {
    "tcp-socket": {
      "command": "TcpSocketMCP",
      "env": {}
    }
  }
}

Option 2: From source

{
  "mcpServers": {
    "tcp-socket": {
      "command": "python",
      "args": ["/path/to/tcp-socket-mcp/run.py"],
      "env": {}
    }
  }
}

Development Setup

# Clone the repository
git clone https://github.com/kaseyk/tcp-socket-mcp.git
cd tcp-socket-mcp

# Install with uv (recommended)
uv pip install -e .

# Or install with pip
pip install -e .

# Run the server directly
python run.py

# Or use the command
TcpSocketMCP

Available Tools

Once configured via MCP, the following tools become available to the AI model:

Core Connection Tools

tcp_connect

Opens a TCP connection to any host:port

• Returns a connection_id for subsequent operations
• Supports custom connection_id for pre-registered triggers
• Example: tcp_connect("example.com", 80)

tcp_send

Sends data over an established connection

• Encoding options: utf-8, hex (recommended for binary), base64
• Hex format: Plain pairs like "48656C6C6F" for "Hello"
• Terminator: Optional hex suffix like "0D0A" for CRLF

tcp_read_buffer

Reads received data from connection buffer

• Data may not be immediately available after sending
• Buffer stores all received data until cleared
• Supports index/count for partial reads
• Format options: utf-8, hex, base64

tcp_disconnect

Closes connection and frees resources

• Always close connections when done
• All triggers automatically removed

Advanced Features

tcp_set_trigger

Sets automatic responses for pattern matches

• Pre-registration: Set triggers before connecting for immediate activation
• Supports regex patterns with capture groups ($1, $2)
• Response fires automatically when pattern matches
• Perfect for protocol handshakes (IRC PING/PONG, etc.)

tcp_connect_and_send

Combines connect + send in one atomic operation

• Essential for time-sensitive protocols
• Useful for immediate handshakes or banner grabbing
• Returns connection_id for further operations

Utility Tools

• tcp_list_connections: View all active connections with statistics
• tcp_connection_info: Get detailed info about specific connection
• tcp_buffer_info: Check buffer statistics without reading data
• tcp_clear_buffer: Clear received data from buffer
• tcp_remove_trigger: Remove specific auto-response trigger

Usage Examples

Basic TCP Communication

# Connect to a service
conn_id = tcp_connect("example.com", 80)

# Send data (hex encoding recommended for protocols)
tcp_send(conn_id, "474554202F20485454502F312E310D0A", encoding="hex")  # GET / HTTP/1.1\r\n

# Read response (may need to wait for data)
response = tcp_read_buffer(conn_id)

# Clean up
tcp_disconnect(conn_id)

Automated Protocol Handling

# Pre-register trigger for IRC PING/PONG
tcp_set_trigger("irc-conn", "ping-handler", "^PING :(.+)", "PONG :$1\r\n")

# Connect with pre-registered triggers
tcp_connect("irc.server.com", 6667, connection_id="irc-conn")
# PING responses now happen automatically!

Working with Binary Protocols

# Use hex encoding for precise byte control
tcp_send(conn_id, "0001000400000001", encoding="hex")  # Binary protocol header

# Read response in hex for analysis
response = tcp_read_buffer(conn_id, format="hex")

Important Notes

Hex Encoding for Line Endings

Many text protocols (HTTP, SMTP, IRC) require specific line endings. Use hex encoding to avoid JSON escaping issues:

# Common hex sequences:
# 0D0A     = \r\n (CRLF) - HTTP, SMTP, IRC
# 0A       = \n (LF) - Unix line ending
# 0D0A0D0A = \r\n\r\n - HTTP header terminator
# 00       = Null byte - Binary protocols

Timing Considerations

• Network responses aren't instant - use tcp_buffer_info to check for data
• Consider implementing retry logic with small delays
• Buffer accumulates all received data - clear when needed

🧪 Testing & Quality

TcpSocketMCP maintains enterprise-grade quality through comprehensive testing:

Test Coverage

• 85% Coverage (exceeds 80% target)
• 80+ Comprehensive Tests across all components
• Cross-Platform Testing (Ubuntu, Windows, macOS)
• Python 3.10-3.12 Support

Quality Gates

• Automated CI/CD with GitHub Actions
• Security Scanning with Bandit and Safety
• Code Quality Analysis with Ruff and MyPy
• Performance Monitoring and complexity analysis

Running Tests Locally

# Install with test dependencies
uv pip install -e .
uv pip install pytest pytest-asyncio pytest-cov

# Run full test suite with coverage
uv run pytest tests/ --cov=src/TcpSocketMCP --cov-report=term-missing

# Quick test run
uv run pytest

See TESTING.md for comprehensive testing documentation.

License

MIT