Qwen Code Setup

Connect QCode to the Qwen Code CLI with modelProviders in ~/.qwen/settings.json: Anthropic and OpenAI-compatible base URLs, /model, /doctor

Need an API key? Sign up →·Endpoints & base URLs →

Last verified: 2026-10-13 · 📄 Per official docs (Qwen Code 0.25.1)

At a glance¶

Qwen Code is an open-source terminal coding assistant, published on npm as @qwen-code/qwen-code; this page was checked against version 0.25.1. Its documentation sorts connecting "third-party providers such as OpenAI, Anthropic, Google, Azure OpenAI, OpenRouter, Requesty, ModelScope, or a self-hosted endpoint" under API Key authentication — that is the endpoint route QCode uses. The modelProviders setting declares which models exist and how to reach them, and /model switches between them at runtime.

Which protocol¶

Qwen Code documents the protocol as part of the provider key: for built-in providers the key must be a valid auth type (openai, anthropic, gemini, vertex-ai). Each model entry is then identified by its protocol, its id, and its configured baseUrl.

🔴 Never call Claude through the OpenAI-compatible base URL. QCode's OpenAI endpoint rejects Claude model ids; the response reads Model '<id>' is not available on this endpoint. Claude is configured on the anthropic side with https://api.qcode.cc/api. GPT ids are not served on the Anthropic side either.

Type the base URL exactly as above: no trailing slash, and do not put the request path into the field — you enter the base, not .../v1/messages or .../chat/completions. If you use a regional node, the Anthropic base becomes https://asia.qcode.cc/api or https://us.qcode.cc/api; the key is the same one.

Install¶

npm installation requires Node.js 22 or newer:

npm install -g @qwen-code/qwen-code@latest

On macOS and Linux you can use Homebrew instead:

brew install qwen-code

Then open a terminal in your project and start it:

cd /path/to/your-project
qwen

Type /help in the welcome screen for the list of commands.

Configure Claude (Anthropic side)¶

Qwen Code's recommended path is a single file, ~/.qwen/settings.json. Every key below is taken verbatim from its documentation: modelProviders holds one array per auth type, an entry needs an id, and name, baseUrl and generationConfig are optional; envKey names the environment variable that carries the credential.

{
  "modelProviders": {
    "anthropic": [
      {
        "id": "claude-sonnet-5",
        "name": "claude-sonnet-5",
        "envKey": "ANTHROPIC_API_KEY",
        "baseUrl": "https://api.qcode.cc/api"
      }
    ]
  },
  "env": {
    "ANTHROPIC_API_KEY": "cr_your_qcode_key"
  },
  "security": {
    "auth": {
      "selectedType": "anthropic"
    }
  },
  "model": {
    "name": "claude-sonnet-5"
  }
}

What the four blocks do, in the words of the official table: modelProviders declares which models are available and how to connect to them; env stores API keys in settings.json as a fallback, with shell export and .env files taking precedence; security.auth.selectedType tells Qwen Code which protocol to use on startup; model.name is the default model activated at startup and must match one of your id values.

The Anthropic protocol also has an environment-variable form — Qwen Code lists ANTHROPIC_API_KEY, ANTHROPIC_BASE_URL and ANTHROPIC_MODEL for it:

export ANTHROPIC_API_KEY="cr_your_qcode_key"
export ANTHROPIC_BASE_URL="https://api.qcode.cc/api"
export ANTHROPIC_MODEL="claude-sonnet-5"
qwen

You can also run /auth inside the CLI: its third top-level option, Custom Provider, is the documented entry for manually connecting "a local server, proxy, or unsupported provider", and it supports OpenAI, Anthropic, Gemini and other compatible endpoints.

About the Anthropic path. The official Anthropic examples write baseUrl down to a /v1 suffix ("baseUrl": "https://api.anthropic.com/v1"), and the documentation does not state how the Anthropic endpoint builds the request path from a base. So this page only states QCode's documented value: enter https://api.qcode.cc/api and nothing more. We do not claim a tested outcome for any other spelling.

Configure GPT and Chinese models (OpenAI-compatible side)¶

The openai auth type is documented as covering not only OpenAI's own API but also any OpenAI-compatible endpoint, including aggregated providers — which is the endpoint QCode's GPT and Chinese models are served on. The official rule for a hosted gateway applies here: set baseUrl to the API's /v1 root rather than the full chat path, because the SDK appends the request path itself. https://api.qcode.cc/openai/v1 is that root.

{
  "modelProviders": {
    "openai": [
      {
        "id": "gpt-6.1-sol",
        "name": "gpt-6.1-sol",
        "envKey": "OPENAI_API_KEY",
        "baseUrl": "https://api.qcode.cc/openai/v1"
      },
      {
        "id": "deepseek-v4.1-flash",
        "name": "deepseek-v4.1-flash",
        "envKey": "OPENAI_API_KEY",
        "baseUrl": "https://api.qcode.cc/openai/v1"
      }
    ]
  },
  "env": {
    "OPENAI_API_KEY": "cr_your_qcode_key"
  },
  "security": {
    "auth": {
      "selectedType": "openai"
    }
  },
  "model": {
    "name": "gpt-6.1-sol"
  }
}

Only /model exposes non-default auth types, so add the models you want to the array and pick them there. Model ids come from the QCode catalog: gpt-6.1-sol, gpt-6-sol, gpt-5.6-terra, or a Chinese id such as glm-5.3, deepseek-v4.1-flash or qwen3.8-max.

The environment-variable form uses the three variables Qwen Code lists for this protocol:

export OPENAI_API_KEY="cr_your_qcode_key"
export OPENAI_BASE_URL="https://api.qcode.cc/openai/v1"
export OPENAI_MODEL="gpt-6.1-sol"
qwen

Verify¶

Ask the CLI what it sees:

/doctor
/model

/doctor is the documented way to check your current configuration at any time; in the quickstart command table its line reads "Check current authentication and environment". /model opens the picker over the entries you declared, and switching takes effect without a restart — modelProviders edits in settings.json are picked up by a running interactive session, but you need to reopen /model to see them.

Then check the endpoint itself, independently of the CLI. The Anthropic Messages request carries the key in x-api-key and a version in anthropic-version:

KEY="cr_your_qcode_key"
curl -X POST https://api.qcode.cc/api/v1/messages \
  -H "x-api-key: $KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model":"claude-sonnet-5","max_tokens":32,"messages":[{"role":"user","content":"ping"}]}'

This page was assembled from documentation, and we hold no test key, so no response from that request is claimed here. If curl works and the CLI does not, the difference is in your ~/.qwen/settings.json; if curl fails, the base URL or the key is the problem — continue with Troubleshooting.

Troubleshooting / Known limitations¶

Further points worth knowing:

• Gemini: Qwen Code can be configured for the gemini auth type, but QCode does not offer the Gemini series — use Claude, GPT, or a Chinese model.
• Not on offer: QCode does not offer gpt-6-luna or gpt-5.6-luna; choose another GPT model such as gpt-6.1-sol, gpt-6-sol or gpt-5.6-terra. Which model fits which task is covered in Choosing a model.
• Restart semantics: modelProviders changes are picked up by a running session, while providerProtocol is read once at startup and requires a restart.
• No credential flags on this page: the official pages spell the OpenAI base-URL flag two different ways, so configure credentials only through ~/.qwen/settings.json or environment variables.
• Base URL hygiene: no trailing slash, and never type the request path into baseUrl.
• Cost: billing is per token; per-model rates are listed on qcode.cc/models. This page states no prices.

Related¶

• Endpoints and API paths — which protocol serves which model family
• Tool compatibility — other coding tools that speak Anthropic or OpenAI-compatible protocols
• OpenCode setup — another terminal CLI on the same two endpoints
• Chinese models
• Choosing a model
• Troubleshooting

Enterprise →