Setup and API reference

Step 1: Check your version

You need 2.1.129 or newer. The CLI and the VS Code extension share ~/.claude/settings.json.

claude --version

claude update
# or, if you installed it with npm:
npm install -g @anthropic-ai/claude-code

Step 2: Point it at the gateway (terminal / CLI)

These apply to the current terminal only — nothing permanent, undone by closing the window. Replace YOUR_SERVIQO_KEY with your key.

export ANTHROPIC_BASE_URL=https://api.serviqo.ai
export ANTHROPIC_AUTH_TOKEN=YOUR_SERVIQO_KEY

claude

ANTHROPIC_BASE_URL

Sends requests to Serviqo instead of Anthropic. No /v1 on the end.

ANTHROPIC_AUTH_TOKEN

Your key. Sent as the Authorization header on every request.

Step 3: VS Code extension — User settings.json

The Claude Code extension does not read your shell profile. Open Ctrl+Shift+P → Preferences: Open User Settings (JSON), merge this in, replace the key, save, then Developer: Reload Window. It must be an array of name/value objects — a plain object crashes the extension.

Only these two variables. Do not add ANTHROPIC_DEFAULT_OPUS_MODEL (or Sonnet / Haiku / Fable) and do not set CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY — those make /model say “Custom Opus model” instead of the Max labels.

{
  "claudeCode.preferredLocation": "panel",
  "claudeCode.environmentVariables": [
    {
      "name": "ANTHROPIC_BASE_URL",
      "value": "https://api.serviqo.ai"
    },
    {
      "name": "ANTHROPIC_AUTH_TOKEN",
      "value": "YOUR_SERVIQO_KEY"
    }
  ]
}

Step 4: Max-style /model picker — ~/.claude/settings.json

This is the file Claude Code uses for the picker. It keeps the key off your bashrc, and modelPicker copies the official Max rows: Default, Opus (1M context), Fable, Sonnet, Haiku.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.serviqo.ai",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_SERVIQO_KEY"
  },
  "model": "opus[1m]",
  "modelPicker": {
    "replaceBuiltInOptions": true,
    "options": [
      {
        "model": "opus[1m]",
        "label": "Opus (1M context)",
        "description": "Opus 5 with 1M context · Best for everyday, complex tasks"
      },
      {
        "model": "fable",
        "label": "Fable",
        "description": "Fable 5.1 · Most capable for your hardest and longest-running tasks · $10/$50 per Mtok"
      },
      {
        "model": "sonnet",
        "label": "Sonnet",
        "description": "Sonnet 5 · Efficient for routine tasks"
      },
      {
        "model": "haiku",
        "label": "Haiku",
        "description": "Haiku 4.5 · Fastest for quick answers"
      }
    ]
  }
}

Step 5: Optional: shell profile (CLI only)

cat >> ~/.bashrc <<'SERVIQO'

# Serviqo
export ANTHROPIC_BASE_URL=https://api.serviqo.ai
export ANTHROPIC_AUTH_TOKEN=YOUR_SERVIQO_KEY
SERVIQO

source ~/.bashrc

Skip this if you use the VS Code extension and want to keep a Max login as the default on this machine. Shell exports override Max for every Claude Code process started from that shell.

set -gx ANTHROPIC_BASE_URL https://api.serviqo.ai
set -gx ANTHROPIC_AUTH_TOKEN YOUR_SERVIQO_KEY

[Environment]::SetEnvironmentVariable('ANTHROPIC_BASE_URL', 'https://api.serviqo.ai', 'User')
[Environment]::SetEnvironmentVariable('ANTHROPIC_AUTH_TOKEN', 'YOUR_SERVIQO_KEY', 'User')

Step 6: Choose a model

Start Claude Code and type /model. You should see the same rows as Max: Default (recommended), Opus (1M context), Fable, Sonnet, Haiku. The pill shows Opus 5 (1M) when that row is selected. Pick a row; do not type a model id by hand. The gateway accepts the official Anthropic names those rows send.

Step 1: Install the extension

In VS Code — or VSCodium, or Windsurf — open the Extensions view with Ctrl+Shift+X (Cmd+Shift+X on macOS), search for Cline or Roo Code, and install one of them. Then click its icon in the activity bar to open the panel.

Step 2: Enter the settings

Open Settings — the gear at the top of the extension panel — and enter exactly this.

Then click Done. If your version shows an optional Model Configuration section — context window, max output tokens, image support — leaving it at the defaults is fine. If you would rather set it exactly, the numbers are in the models section.

Step 3: Say hello

In the extension's chat box, type:

Read the files in this folder and tell me what this project does.

If you get an answer, you are done.

Step 1: Open Cursor's model settings

Cmd+Shift+J on macOS, Ctrl+Shift+J on Windows and Linux, then Models in the sidebar. This is Cursor's own settings window, not VS Code's. If the shortcut does nothing on your build, run Cursor Settings from the command palette. Scroll to API Keys.

Step 2: Enter the key and the base URL

Step 3: Add the model ids by hand

Cursor does not ask us which models you may use. On the same Models page click + Add Model and add each model as its own entry, spelled exactly — all lowercase, single hyphen, no spaces.

claude-haiku-5
claude-sonnet-5
claude-opus-5
claude-fable-5-1

Add only the models your key includes; one you are not entitled to produces an error the first time you select it. This lists exactly which those are:

curl -s https://api.serviqo.ai/v1/models \
  -H "Authorization: Bearer YOUR_SERVIQO_KEY"

Step 4: Verify it really is going through the gateway

Do not assume. Cursor's failure mode is to answer silently from one of its own models. Open Chat (Cmd/Ctrl+L), select a model in the model picker, and send something trivial:

Reply with the single word: routed.

Then open the key status page and look at the spend. Spend is recorded a moment after a request finishes, so give it up to a minute. If the reply arrives but spend does not move, the request did not go through us.

A check that takes Cursor out of the picture entirely:

curl -s https://api.serviqo.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_SERVIQO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "messages": [
      {"role": "user", "content": "Reply with the single word: routed."}
    ],
    "max_tokens": 32
  }'

A reply here means the key and the gateway are fine and anything still wrong is in Cursor's settings. An error here names the actual problem.