Installation & setup

This guide helps you set up the environment Noxia needs on macOS or Windows. The example below uses Claude Code as the agent tool and takes about 5 minutes from start to finish.

New to the terminal? We recommend the CCSwitch GUI tool, which lets you complete steps 2 and 5 below without typing any commands.

Prerequisites

  • A Noxia account (used to authorize access in step 6).
  • Node.js 18 or later installed (needed for npx when installing the Skills in step 4; also needed if you install Claude Code via npm).

1. Install the agent tool (Claude Code)

macOS / Linux:

curl -fsSL https://claude.ai/install.sh | bash

Windows (PowerShell):

irm https://claude.ai/install.ps1 | iex

Or use npm (cross-platform):

npm install -g @anthropic-ai/claude-code

Once installed, verify the version:

claude --version

2. (Optional) Configure a custom model API

If you don’t have a Claude subscription, you can connect Claude Code to a third-party LLM that is compatible with the Anthropic API (such as DeepSeek, Qwen, and others). We recommend doing this in Claude Code’s config file: edit ~/.claude/settings.json (on Windows, C:\Users\<username>\.claude\settings.json; create the file if it doesn’t exist) and, under env, fill in the endpoint, key, and model names your provider supplies.

Using DeepSeek as an example:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-API-Key",
    "ANTHROPIC_MODEL": "deepseek-v4-pro",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-v4-pro",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-v4-pro",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-v4-flash"
  }
}

What each field means:

  • ANTHROPIC_BASE_URL: your provider’s Anthropic-compatible endpoint.
  • ANTHROPIC_AUTH_TOKEN: the API key you obtained from that provider.
  • ANTHROPIC_DEFAULT_OPUS_MODEL / ANTHROPIC_DEFAULT_SONNET_MODEL / ANTHROPIC_DEFAULT_HAIKU_MODEL: map Claude’s Opus / Sonnet / Haiku tiers to your provider’s specific models; Claude Code automatically picks the matching tier based on task complexity.

For the exact endpoint, key, and available model names, refer to your provider’s documentation (see the Alibaba Cloud Model Studio integration example). Save the file and restart claude for the changes to take effect.

Don’t want to edit the config file by hand? CCSwitch provides a GUI for switching between model providers with one click.

3. Start Claude Code

Open a terminal, change into your project folder, and start it:

cd /path/to/your-project
claude

To skip per-action permission prompts during a session, add this flag (use it only in projects you trust):

claude --dangerously-skip-permissions

4. Install the Noxia companion Skills

In your project root, use the skills tool to install the Noxia companion Skills in one step:

npx skills add muwenbo/noxia-skills -a claude-code

This command installs all of the Skills from the noxia-skills repository into the .claude/skills/ directory, and Claude Code discovers them automatically. See the Skills overview for what each Skill does and how they fit the workflow.

5. Configure the Noxia MCP service

Noxia exposes its capabilities as an MCP service over Streamable HTTP. In your project directory, run:

claude mcp add --transport http noxia https://api.viknox.com/mcp

You can also simply tell Claude Code in the session:

Set up a Streamable HTTP-based Noxia MCP service in this project, at https://api.viknox.com/mcp.

6. Authorize access to your Noxia account

In the Claude Code session, type:

/mcp

Select noxia → Authenticate in turn. A browser authorization window will open—click Allow to grant Claude Code access to your Noxia account.

7. Get started

Your environment is ready. Now head to the Playground and start your first variant analysis with the sample data we provide.