mcp-conserver

MCP server for interacting with serial consoles managed by a conserver daemon.

Allows AI agents (e.g. Claude) to connect to serial consoles, send commands, read output, and manage console sessions through the Model Context Protocol.

Features

• Connect to a console in attach, spy (read-only), or force mode
• Send commands and read output interactively
• Send control characters (Ctrl-C, Ctrl-D, etc.)
• Query who is connected to consoles
• Disconnect users from consoles
• SSL/TLS and GSSAPI/Kerberos authentication

Installation

python -m venv venv
source venv/bin/activate
pip install -e .

Configuration

The server reads ~/.consolerc for default master server, port, and username:

config * {
  master conserver.example.com;
  port 782;
  username myuser;
}

All defaults can be overridden per tool call.

Usage with Claude Code

Activate the virtual environment before starting Claude Code:

source /path/to/mcp-conserver/venv/bin/activate
claude

Then add to .mcp.json in your project directory (see mcp.json.example):

{
  "mcpServers": {
    "conserver": {
      "command": "mcp-conserver"
    }
  }
}

Tools

Architecture

Uses libconserver.py, a pure Python implementation of the conserver client protocol, supporting SSL/TLS (anonymous DH), GSSAPI/Kerberos authentication, and multi-hop server redirects.

Each console session runs a background reader thread that continuously drains the socket into a buffer, ensuring no data is lost between tool calls.

License

GPLv2 - see LICENSE.