OpenCode

Connect OpenCode to 魔方AI with stored credentials and an OpenAI-compatible custom provider.

Before you begin

Create an active 魔方AI token and note the exact model IDs available to it.
This guide uses OpenAI Chat Completions with the base URL https://www.mofangai001.cn/v1.
Windows users can install with npm, Chocolatey, or Scoop; OpenCode officially recommends WSL for the best Windows experience.

Configuration steps

1. Install OpenCode

On macOS or Linux, use the official installer:

curl -fsSL https://opencode.ai/install | bash

You can also install OpenCode with npm, Bun, or pnpm:

npm install -g opencode-ai

bun install -g opencode-ai

pnpm install -g opencode-ai

Run opencode to open the terminal UI after installation.

2. Save the 魔方AI credential

In OpenCode, enter:

/connect

Scroll to the bottom of the provider list, choose Other, and enter:

Field	Value
Provider ID	`mofang`
API Key	`sk-your-api-key`

The provider ID must exactly match the key used in opencode.json in the next step. /connect only stores the credential in ~/.local/share/opencode/auth.json; it does not configure the 魔方AI endpoint or models.

3. Configure the 魔方AI provider

For all projects, add the provider to ~/.config/opencode/opencode.json. For the current project only, create opencode.json in the project root. Project configuration has higher precedence than global configuration.

{
    "$schema": "https://opencode.ai/config.json",
    "provider": {
        "mofang": {
            "npm": "@ai-sdk/openai-compatible",
            "name": "魔方AI",
            "options": {
                "baseURL": "https://www.mofangai001.cn/v1"
            },
            "models": {
                "your-model-id": {
                    "name": "Your Model"
                }
            }
        }
    }
}

Replace your-model-id with the complete model ID shown in 魔方AI. @ai-sdk/openai-compatible uses /v1/chat/completions. If a model explicitly requires /v1/responses, use @ai-sdk/openai instead; do not mix the two API formats.

4. Select a 魔方AI model

Restart OpenCode and enter:

/models

Choose your-model-id under 魔方AI. The provider ID, credential ID, and the mofang key in the configuration must all match.

Verify the connection

First confirm that the credential is stored:

opencode auth list

After mofang appears in the list, start OpenCode, select the 魔方AI model, and send a test request. A normal response that can use the project context confirms the connection.

Troubleshooting

Why is 魔方AI missing after I run /connect?

/connect only stores a credential. You must also add provider.mofang, the base URL, and the model list to a global or project opencode.json, then restart OpenCode.

Why does OpenCode report a missing credential?

Make sure the Provider ID entered under /connect exactly matches mofang in the configuration, and run opencode auth list to confirm the credential was saved.

Why does the base URL include /v1?

@ai-sdk/openai-compatible expects an OpenAI-compatible base URL and calls Chat Completions beneath it. The 魔方AI base URL is https://www.mofangai001.cn/v1.

Why is the model missing from /models?

Check the model ID in the models object, the JSON syntax, and the configuration file location. The model ID must exactly match a model available to your 魔方AI token.

Security

Use /connect to store credentials instead of placing a real token in project configuration that may be committed. Do not share auth.json, configuration screenshots, or logs. Create a dedicated OpenCode token with model, quota, and IP restrictions.