TcpSocketMCP is an MCP server that gives AI models direct access to raw TCP sockets, letting them interact with network services without an HTTP wrapper. It supports multiple simultaneous connections, response data buffering, and automatic triggers. Eleven tools are exposed: tcp_connect opens a connection to any host:port and returns a connection_id for subsequent operations; tcp_send sends data through an established connection with a choice of encoding (utf-8, hex, or base64); tcp_read_buffer reads received data with support for partial reads and formatting options; tcp_disconnect closes a connection, frees its resources, and automatically removes any triggers attached to it; tcp_set_trigger sets an automatic response for pattern matches, supporting regex patterns with capture groups; tcp_connect_and_send combines connect and send into a single atomic operation for time-sensitive protocols; tcp_list_connections lists all active connections with statistics; tcp_connection_info gets detailed information about one connection; tcp_buffer_info checks buffer statistics without consuming the data; tcp_clear_buffer clears received data from a connection's buffer; and tcp_remove_trigger removes a specific auto-response trigger.

Use this for direct interaction with embedded devices and IoT systems, debugging and testing network protocols, integrating with legacy systems that have no HTTP wrapper, protocol analysis and reverse engineering, or setting up automatic pattern-based responses for protocols like IRC or telnet (for example, automatic PING/PONG replies). It also fits tasks like polling unknown devices for identification or sending raw HTTP requests to a web server over a manually managed socket. Because network responses don't arrive instantly, the source recommends polling tcp_buffer_info to check for available data rather than assuming a response is immediately present after sending.

Hex encoding is recommended for binary protocols and for precise control over line endings, with common sequences called out explicitly: 0D0A for CRLF, 0A for LF, and 00 for null bytes. The server supports binary protocols via hex encoding, regex-based pattern matching for automatic responses, and is cross-platform compatible across Ubuntu, Windows, and macOS. The project maintains 85% test coverage with CI/CD and quality assurance in place.

Installation is available via PyPI:

pip install TcpSocketMCP

or via uv add TcpSocketMCP, or from source by cloning the repository and running uv pip install -e . followed by python run.py for direct execution. For Claude Desktop, an installed package is configured with command: TcpSocketMCP, while a from-source install is configured with command: python and args pointing at the local run.py path.

Network protocol developers, embedded/IoT engineers, and security researchers who need an AI assistant to open raw TCP connections, send and receive data in multiple encodings, and set up pattern-triggered automatic responses for protocol testing or reverse engineering.