Setup #

Start the MCP server with:

quantum-cli mcp serve

Explore the MCP with npx @modelcontextprotocol/inspector quantum-cli mcp serve

Agent Configuration #

The MCP works using stdio, which is the default transport when you add it to your agent.

Claude #

claude mcp add --scope user \
  quantum-cli -- quantum-cli mcp serve

Codex #

codex mcp add --scope user \
  quantum-cli -- quantum-cli mcp serve

JSON Configuration #

For agents that use JSON configuration files (e.g., Claude Desktop, VS Code extensions):

{
  "mcpServers": {
    "quantum-cli": {
      "command": "quantum-cli",
      "args": ["mcp", "serve"]
    }
  }
}

Enabling Destructive Operations #

By default, the MCP server is read-only. To enable tools that can modify your infrastructure, use the --enable-danger flag:

claude mcp add --scope user \
  quantum-cli -- quantum-cli mcp serve --enable-danger

The --enable-danger flag enables operations like deployments, deletions, and restarts. Only enable this if you trust your AI agent to perform these actions. Or make it project-scoped by omitting --scope user.

For JSON configuration:

{
  "mcpServers": {
    "quantum-cli": {
      "command": "quantum-cli",
      "args": ["mcp", "serve", "--enable-danger"]
    }
  }
}