CCS

How the integration works

This guide uses the official CCS OpenAI-Compatible Routing path:

Claude Code
    ↓ Anthropic Messages
Local CCS proxy (127.0.0.1, port selected by CCS)
    ↓ OpenAI Chat Completions
https://www.mofangai001.cn/v1/chat/completions

Use this 魔方AI API Base URL:

https://www.mofangai001.cn/v1

Do not enter the complete /chat/completions request URL or add a second /v1. CCS starts the local proxy, translates the protocol, and forwards the request to 魔方AI.

Prerequisites

Before you begin, make sure that:

• Node.js 18 or later is installed.
• Claude Code is installed and claude --version works in your terminal.
• You have signed in to 魔方AI and created an active token on the Token Management page.
• You know the exact model IDs allowed by the token. Each ID must match the value shown in 魔方AI.

1. Install CCS

Run this in PowerShell, Windows Terminal, or a macOS/Linux terminal:

npm install -g @kaitranntt/ccs@latest

Verify the installation:

ccs --version

2. Open API Profiles

Run:

ccs config

CCS opens its local configuration dashboard in your browser. Select API Profiles in the sidebar, then click New or Create Profile.

3. Create the 魔方AI profile

Enter these values under Basic Information:

Field	Value
Profile Name	mofang
API Base URL	https://www.mofangai001.cn/v1
API Key	Your 魔方AI token, for example sk-your-api-key
Default Target	Claude Code

The profile name becomes part of the launch command. Lowercase letters, numbers, and hyphens are recommended.

4. Configure model mappings

Open Model Configuration and enter model IDs allowed by the current 魔方AI token:

Field	Purpose
Default Model	The default model, for example your-model-id
Opus Model	Model used when Claude Code requests the Opus tier
Sonnet Model	Model used when Claude Code requests the Sonnet tier
Haiku Model	Model used when Claude Code requests the Haiku tier

If you only need one model, use the same exact model ID for all four fields. For separate mappings, verify that every ID is allowed by the token.

Enable the CCS [1m] option only when the selected model and upstream account actually support long context. The option cannot bypass upstream entitlements, quota, or billing requirements.

Save the profile when finished.

5. Confirm the OpenAI-compatible provider type

This check is required for 魔方AI. Open the new mofang profile and confirm that the environment variables or Additional Variables contain:

CCS_DROID_PROVIDER=generic-chat-completion-api

This tells CCS that 魔方AI exposes OpenAI Chat Completions and that Claude Code traffic must go through the local protocol-conversion proxy.

If your CCS dashboard does not show this field, edit the profile settings file directly:

• Windows: %USERPROFILE%\.ccs\mofang.settings.json
• macOS/Linux: ~/.ccs/mofang.settings.json

The important fields should look like this:

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://www.mofangai001.cn/v1",
    "ANTHROPIC_AUTH_TOKEN": "sk-your-api-key",
    "ANTHROPIC_MODEL": "your-model-id",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "your-model-id",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "your-model-id",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "your-model-id",
    "CCS_DROID_PROVIDER": "generic-chat-completion-api"
  }
}

6. Launch Claude Code

From the project directory where you want to work, run:

ccs mofang

CCS starts the profile's local OpenAI-compatible proxy and launches Claude Code. The terminal normally prints a message similar to:

Using local OpenAI-compatible proxy for "mofang" on port ...

Check the proxy from another terminal if needed:

ccs proxy status mofang

Send a short test message in Claude Code and confirm that the model returns a response.

Optional: create the profile in the terminal

Instead of the dashboard, run the interactive wizard:

ccs api create mofang

Enter the following when prompted:

1. API Base URL: https://www.mofangai001.cn/v1
2. API Key: your 魔方AI token
3. Default Model: a model ID allowed by the token
4. Opus / Sonnet / Haiku mappings: configure as needed
5. Default Target: choose Claude Code, not Factory Droid

After the wizard completes, still verify CCS_DROID_PROVIDER as described above. Avoid putting the real token in a command-line --api-key argument because it can remain in shell history, process details, or recordings.

Configuration files and security

CCS stores its main configuration under ~/.ccs/. Profile settings files contain the complete 魔方AI token.

• Do not share *.settings.json, exported profiles, terminal screenshots, or debug logs.
• Do not commit ~/.ccs/ to Git.
• Use a dedicated 魔方AI token for CCS so model access, quota, IP rules, and revocation can be managed independently.
• If a token leaks, disable or delete it in 魔方AI immediately and update the CCS profile.
• The local CCS proxy binds to 127.0.0.1 by default. Do not expose it to a LAN or the public internet without reliable authentication, firewall rules, and a trusted network.

Troubleshooting

Official resources

• CCS website
• CCS documentation
• CCS GitHub repository
• CCS OpenAI-Compatible Provider Routing
• CCS package on npm