Integrating with Claude Code

Last Updated: 2026-05-13 Supported Model: intern-s2-preview API Endpoint: https://chat.intern-ai.org.cn/v1/messages

Our Claude-like API is compatible with the Claude Code client, so with intern-s2-preview you can use Intern directly inside Claude Code. There are three ways to wire it up — pick whichever feels best:

Method 1 · Environment variables: one-shot, doesn't touch any existing config.
Method 2 · settings.json: persistent, written into Claude Code's own config.
Method 3 · Shell alias: lives alongside the official Claude, switch on demand.

Getting Ready

Make sure Claude Code is installed (recommended >= 1.0.0):

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

Then grab an API token from the Intern console: API Tokens. Every method below needs it.

Under the hood, you only need to tell Claude Code two things: where to send the request and who's asking. These map to ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. All three methods just write them in different places.

Method 1: Environment Variables (fastest, try it first)

Set them in the current shell — nothing else changes. Close the terminal and everything is back to normal. Great for "let me just see if it works".

Linux / macOS

Open your terminal and paste these four lines (swap in your token):

export ANTHROPIC_BASE_URL="https://chat.intern-ai.org.cn"
export ANTHROPIC_AUTH_TOKEN="your-api-token"
export ANTHROPIC_MODEL="intern-s2-preview"
export ANTHROPIC_SMALL_FAST_MODEL="intern-s2-preview"

Then launch Claude Code:

claude

Inside Claude Code, run /status. If you see our Base URL and intern-s2-preview as the model, you're good 🎉

Windows PowerShell

$env:ANTHROPIC_BASE_URL = "https://chat.intern-ai.org.cn"
$env:ANTHROPIC_AUTH_TOKEN = "your-api-token"
$env:ANTHROPIC_MODEL = "intern-s2-preview"
$env:ANTHROPIC_SMALL_FAST_MODEL = "intern-s2-preview"

claude

To make it survive restarts, set them at the user level:

[Environment]::SetEnvironmentVariable("ANTHROPIC_BASE_URL", "https://chat.intern-ai.org.cn", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_AUTH_TOKEN", "your-api-token", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_MODEL", "intern-s2-preview", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_SMALL_FAST_MODEL", "intern-s2-preview", "User")

Windows CMD

set ANTHROPIC_BASE_URL=https://chat.intern-ai.org.cn
set ANTHROPIC_AUTH_TOKEN=your-api-token
set ANTHROPIC_MODEL=intern-s2-preview
set ANTHROPIC_SMALL_FAST_MODEL=intern-s2-preview

claude

Making it Permanent on Linux / macOS

The export lines above only live as long as the terminal does. To have them load every time, add them to your shell startup file.

Step 1 — figure out which shell you're using:

echo $SHELL

Output mentions zsh → edit ~/.zshrc (macOS default)
Output mentions bash → edit ~/.bashrc (most Linux distros)

The example below uses zsh; for bash, just swap .zshrc for .bashrc.

Step 2 — open the file with any editor:

# Pick one
nano ~/.zshrc      # Beginner friendly, shortcuts shown at the bottom
vim ~/.zshrc       # If you're a vim person
code ~/.zshrc      # If you have VS Code installed

If the file doesn't exist yet: touch ~/.zshrc first.

Step 3 — append the config at the end of the file:

# Intern Claude Code
export ANTHROPIC_BASE_URL="https://chat.intern-ai.org.cn"
export ANTHROPIC_AUTH_TOKEN="your-api-token"
export ANTHROPIC_MODEL="intern-s2-preview"
export ANTHROPIC_SMALL_FAST_MODEL="intern-s2-preview"

Save and quit (nano: Ctrl+O, Enter, then Ctrl+X; vim: :wq, Enter).

Step 4 — load the changes right now:

source ~/.zshrc

This is equivalent to opening a fresh terminal — the shell re-reads the config. Any new terminal will pick these up automatically.

Method 2: Edit Claude Code's settings.json

⚠️ Heads up: this method overwrites Claude Code's global config — your existing Anthropic key will stop working. If you're actively using the official Claude, prefer Method 1 or Method 3.

Config file location:

Linux / macOS: ~/.claude/settings.json
Windows: %USERPROFILE%\.claude\settings.json

Open it and merge in the env block below (if the file is empty, paste the whole thing as-is):

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://chat.intern-ai.org.cn",
    "ANTHROPIC_AUTH_TOKEN": "your-api-token",
    "ANTHROPIC_MODEL": "intern-s2-preview",
    "ANTHROPIC_SMALL_FAST_MODEL": "intern-s2-preview"
  }
}

Restart Claude Code to apply.

Want to go back to the official Claude? Delete the keys above, or revert settings.json to whatever it used to be. Tip: cp settings.json settings.json.bak before you edit — that's the easy way back.

If you only want this in one project, drop the same config into .claude/settings.json inside the project root. Other projects stay unaffected.

Method 3: Shell alias (coexist with official Claude)

If you want to keep using the official Claude and switch to Intern on demand, this is the cleanest setup: a new command claude-intern, while the original claude stays untouched.

Linux / macOS

Step 1 — open your shell config (same as Method 1 Step 2; zsh users edit ~/.zshrc, bash users edit ~/.bashrc):

nano ~/.zshrc

Step 2 — append the alias:

alias claude-intern='ANTHROPIC_BASE_URL="https://chat.intern-ai.org.cn" \
  ANTHROPIC_AUTH_TOKEN="your-api-token" \
  ANTHROPIC_MODEL="intern-s2-preview" \
  ANTHROPIC_SMALL_FAST_MODEL="intern-s2-preview" \
  claude'

Step 3 — reload:

source ~/.zshrc

Now you have two commands:

claude          # Uses your existing config (e.g. official Claude)
claude-intern   # Routes to Intern intern-s2-preview

Arguments work transparently — claude-intern --help behaves the same as claude --help.

Windows PowerShell

PowerShell aliases can't carry environment variables, so use a function instead — same effect.

Open the profile:

notepad $PROFILE

If it doesn't exist:

New-Item -ItemType File -Path $PROFILE -Force
notepad $PROFILE

Append:

function claude-intern {
    $env:ANTHROPIC_BASE_URL = "https://chat.intern-ai.org.cn"
    $env:ANTHROPIC_AUTH_TOKEN = "your-api-token"
    $env:ANTHROPIC_MODEL = "intern-s2-preview"
    $env:ANTHROPIC_SMALL_FAST_MODEL = "intern-s2-preview"
    claude @args
}

Save, run . $PROFILE to reload, and claude-intern is ready.

Verify

Inside Claude Code, ask anything:

> What model are you? Introduce yourself in one sentence.

Or hit the API directly with curl:

curl -X POST https://chat.intern-ai.org.cn/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: your-api-token" \
  -d '{
    "model": "intern-s2-preview",
    "max_tokens": 128,
    "messages": [{"role": "user", "content": "ping"}]
  }'

A response containing "type": "message" means you're connected.

What Do the Two Model Variables Do?

Easy to be confused about, so here's the breakdown:

Variable	Role	What it affects
`ANTHROPIC_MODEL`	Primary model (everyday chat)	Every normal conversation, coding, reasoning, and tool call you have with Claude Code goes through this model
`ANTHROPIC_SMALL_FAST_MODEL`	Small fast model (background tasks)	Conversation title generation, `/compact` summarization, status-line updates, lightweight classification — anything that needs a quick reply

Claude Code automatically picks the right one for each task — you don't switch manually.

Why we set both to intern-s2-preview? Currently that's the only model we expose. If you set only ANTHROPIC_MODEL, background tasks fall back to the default claude-haiku-*, which doesn't exist on our endpoint — you'll see errors.

A small gotcha about /model: when you use a custom BASE_URL, Claude Code's /model command does not query our model list. It only shows/toggles whatever you set via env vars. To check what's actually live, /status is more reliable — it prints Base URL, Model, and Small Fast Model directly.

FAQ

Q1: claude errors with 401 Unauthorized / Invalid API Key. Double-check ANTHROPIC_AUTH_TOKEN. Common pitfalls: leading/trailing spaces or quotes, an accidental Bearer prefix, expired token.

Q2: /status still shows api.anthropic.com. The variables didn't reach the Claude Code process. Open a fresh terminal, or confirm your ~/.bashrc / ~/.zshrc / $PROFILE was saved and sourced.

Q3: Replies are often empty, or tool calls feel unreliable. For tool-heavy clients like Claude Code, intern-s2-preview must keep thinking_mode enabled (it is by default). Don't disable it via extraBody in your settings.

Q4: What's the max context? intern-s2-preview supports 256K context. Per-request max_tokens should stay ≤ 8K. For big projects, /compact inside Claude Code helps a lot.

Q5: I'm being rate-limited (-20053 / HTTP 429). Default is 30 req/min. Claude Code's multi-agent, tool-heavy flows can blow through that quickly. Request a higher quota at the Rate Limiting Policy page.

Q6: How do I go back to the official Claude?

Method 1: close the terminal, or run unset ANTHROPIC_BASE_URL ANTHROPIC_AUTH_TOKEN ANTHROPIC_MODEL ANTHROPIC_SMALL_FAST_MODEL.
Method 2: edit ~/.claude/settings.json and remove the keys.
Method 3: just use claude — the alias never pollutes the default command.

Q7: Does it work with the VS Code / JetBrains plugin? Yes. The plugins read the same ~/.claude/settings.json, so Method 2 just works. Environment-variable setups only apply to IDE processes launched from that same shell.

Integrating with Claude Code

Getting Ready​

Method 1: Environment Variables (fastest, try it first)​

Linux / macOS​

Windows PowerShell​

Windows CMD​

Making it Permanent on Linux / macOS​

Method 2: Edit Claude Code's settings.json​

Method 3: Shell alias (coexist with official Claude)​

Linux / macOS​

Windows PowerShell​

Verify​

What Do the Two Model Variables Do?​

FAQ​

Getting Ready

Method 1: Environment Variables (fastest, try it first)

Linux / macOS

Windows PowerShell

Windows CMD

Making it Permanent on Linux / macOS

Method 2: Edit Claude Code's settings.json

Method 3: Shell alias (coexist with official Claude)

Linux / macOS

Windows PowerShell

Verify

What Do the Two Model Variables Do?

FAQ