The MCP server connects an AI agent directly to VSNES, allowing it to configure, launch, and monitor satellite network emulations through natural language.

python SatelliteEmulator.py --mcp        # API + MCP
python SatelliteEmulator.py --all        # API + Web + NTP + MCP

Configure the server via environment variables:

export SNES_API_URL=http://localhost:5050   # REST API address
export VM_SSH_USER=debian                   # SSH user for VMs
export VM_SSH_PASS=debian                   # SSH password for VMs

To connect from Claude Desktop, add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "vsnes": {
      "url": "http://localhost:8560/mcp"
    }
  }
}

Typical AI-driven workflow:

1. AI calls get_config_guide to learn what parameters to collect
2. AI interviews the user; if a synthetic constellation is wanted, it calls generate_tle first and passes the resulting path as tle_file
3. AI calls generate_config_toml to write config.toml
4. AI calls generate_docker_compose to create the container manifest
5. AI calls load_scenario → init_scenario → start_simulation to launch the emulation (or prepare_simulation for the one-shot load+init)
6. AI monitors progress with get_emulator_status; at the end of the timeline the run stops but the rules stay applied (start_simulation replays it), and stop_simulation / reset_system tears it down