Documentation

Setup guide

Everything you need to point a client at GrowMoon: install, configure your editor, and call the API. With a key in hand, the whole thing takes about a minute.

Drop-in compatible

No SDK changes

Per-key limits

5-hour rolling window

Built-in tools

Search and image analysis

18 models

One base URL

Prerequisites

Node.js 18 or newer

Download it from nodejs.org

A GrowMoon API key

Get one from GrowMoon or an authorised reseller

A supported client

Claude Code, VS Code, Cursor, Windsurf, Cline or Roo Code

Quick install

The fastest path. The wizard asks for your key, configures the clients you pick, and checks the connection before it exits.

terminal
npx growmoon-cli

What it does

  1. 01Asks for your API key
  2. 02Lets you choose which clients to configure
  3. 03Writes the correct settings for each one
  4. 04Checks the connection

Web search and image analysis need no setup at all: they run server-side.

Windows (PowerShell)

If you would rather run the setup script directly:

powershell
Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser
irm https://growmoon.org/setup.ps1 | iex

macOS and Linux

The shell equivalent:

terminal
curl -fsSL https://growmoon.org/setup.sh | bash

Claude Code CLI

Point the Claude Code CLI at GrowMoon and every model in the lineup becomes selectable from the same session.

Automatic (recommended)

Run npx growmoon-cli and select Claude Code (CLI).

Manual configuration

Create or edit ~/.claude/settings.json:

~/.claude/settings.json
{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "sk-ant-gm-...",
    "ANTHROPIC_BASE_URL": "https://api.growmoon.org",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1",
    "CLAUDE_CODE_AUTO_MODE_SERVER": "0"
  },
  "hasCompletedOnboarding": true
}

Replace sk-ant-gm-... with your full key. No model pins are needed: Claude Code reads the model lineup from GrowMoon. CLAUDE_CODE_AUTO_MODE_SERVER=0 keeps auto mode's checks on your machine, so Claude Code never shows a notice about the gateway.

VS Code

The Claude extension for VS Code reads the same configuration as the CLI.

Automatic (recommended)

Run npx growmoon-cli and select VS Code (Claude Extension).

Manual configuration

Use the same file as the Claude Code CLI above, then restart VS Code once you have saved it.

Cursor

Route Cursor’s AI features through GrowMoon.

Automatic (recommended)

Run npx growmoon-cli and select Cursor.

API routing

Open Settings, then Models, and add an OpenAI-compatible model with:

  • Base URL: https://api.growmoon.org/v1
  • API key: your GrowMoon key (sk-ant-gm-...)
  • Model: claude-sonnet-5

Windsurf

Route Windsurf’s AI provider through GrowMoon.

Automatic (recommended)

Run npx growmoon-cli and select Windsurf.

API routing

Open Settings, then AI Provider, and set the base URL to:

Base URL
https://api.growmoon.org/v1

Cline

Configure the Cline extension for VS Code to use GrowMoon.

Automatic (recommended)

Run npx growmoon-cli and select Cline (VS Code Extension).

Manual configuration

Add to your VS Code settings.json:

settings.json
{
  "cline.apiProvider": "anthropic",
  "cline.anthropicBaseUrl": "https://api.growmoon.org",
  "cline.apiKey": "sk-ant-gm-..."
}

Roo Code

Configure the Roo Code extension for VS Code to use GrowMoon.

Automatic (recommended)

Run npx growmoon-cli and select Roo Code (VS Code Extension).

Manual configuration

Add to your VS Code settings.json:

settings.json
{
  "roo-cline.apiProvider": "anthropic",
  "roo-cline.anthropicBaseUrl": "https://api.growmoon.org",
  "roo-cline.apiKey": "sk-ant-gm-..."
}

Authentication

Every gateway endpoint at https://api.growmoon.org accepts your key through either header:

header options
x-api-key: sk-ant-gm-...
# or
Authorization: Bearer sk-ant-gm-...

Partner dashboard endpoints are separate: they use a session token from the /auth/login endpoint.

Messages

POST/v1/messagesAPI key

Create a message. Anthropic-compatible request and response shapes. Set stream: true for server-sent events.

Request
{
  "model": "claude-sonnet-5",
  "max_tokens": 1024,
  "messages": [
    { "role": "user", "content": "Hello, Claude!" }
  ],
  "stream": false
}
Response
{
  "id": "msg_...",
  "type": "message",
  "role": "assistant",
  "model": "claude-sonnet-5",
  "content": [
    { "type": "text", "text": "Hello! How can I help you today?" }
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 12,
    "output_tokens": 15
  }
}

Streaming. With stream: true you get server-sent events (message_start, content_block_delta and message_stop), forwarded as they arrive.

Models

GET/v1/modelsNone

List every available model. Public, no key required.

Response
{
  "data": [
    {
      "id": "claude-fable-5-2",
      "type": "model",
      "display_name": "Claude Fable 5.2",
      "created_at": "2026-11-09T00:00:00Z",
      "context_window": 1000000
    },
    {
      "id": "claude-fable-5-1",
      "type": "model",
      "display_name": "Claude Fable 5.1",
      "created_at": "2026-09-01T00:00:00Z",
      "context_window": 1000000
    },
    {
      "id": "claude-fable-5",
      "type": "model",
      "display_name": "Claude Fable 5",
      "created_at": "2026-06-09T00:00:00Z",
      "context_window": 1000000
    },
    {
      "id": "claude-opus-5-5",
      "type": "model",
      "display_name": "Claude Opus 5.5",
      "created_at": "2026-09-22T00:00:00Z",
      "context_window": 1000000
    }
  ],
  "has_more": false
}

Truncated above. The live endpoint returns all 18 models; see .

Token counting

POST/v1/messages/count_tokensAPI key

Count the tokens a message would use, without sending it.

Request
{
  "model": "claude-sonnet-5",
  "messages": [
    { "role": "user", "content": "How many tokens is this?" }
  ]
}

Key status

GET/api/key-status?key=None

Read a key's status, 5-hour window usage (as a percentage) and limits. This is what the Check key page calls.

Response
{
  "status": "found",
  "name": "My key",
  "keyPrefix": "sk-ant-gm-Q7xY2kLm9a",
  "isActive": true,
  "rateLimit": 60,
  "unlimited": false,
  "usagePercent": 24.7,
  "windowActive": true,
  "windowStartedAt": "2026-11-17T10:00:00.000Z",
  "windowResetAt": "2026-11-17T15:00:00.000Z",
  "expiresAt": "2026-12-17T00:00:00.000Z",
  "isExpired": false,
  "isOverLimit": false,
  "totalRequests": 142,
  "last24h": { "requests": 28 }
}

Image analysis

POST/tools/understand_imageAPI key

Analyse an image. Accepts an HTTP URL, a local path, or a base64 data URL. 18MB maximum.

Request
{
  "prompt": "Describe what you see in this image",
  "image_url": "https://example.com/photo.jpg"
}

Built-in tools

Both tools run server-side. There is nothing to install on your machine and no MCP server to configure.

Web search

Live web results for up-to-date information. Available automatically in every connected client.

Image analysis

Understands JPEG, PNG and WebP. Works out of the box with no extra configuration.

Available models

Every model below answers on the same key and the same base URL. Pass the ID exactly as written.

Claude Fable 5.2Newclaude-fable-5-21M
Claude Fable 5.1claude-fable-5-11M
Claude Fable 5claude-fable-51M
Claude Opus 5.5Newclaude-opus-5-51M
Claude Opus 5claude-opus-51M
Claude Sonnet 5.5Newclaude-sonnet-5-51M
Claude Sonnet 5claude-sonnet-51M
Claude Haiku 5.5claude-haiku-5-5200K
Claude Opus 4.8claude-opus-4-81M
Claude Opus 4.7claude-opus-4-71M
Claude Opus 4.6claude-opus-4-61M
Claude Sonnet 4.6claude-sonnet-4-61M
Claude Opus 4.5claude-opus-4-5200K
Claude Sonnet 4.5claude-sonnet-4-5-20250929200K
Claude Haiku 4.5claude-haiku-4-5-20251001200K
Claude Opus 4.1claude-opus-4-1-20250805200K
Claude Opus 4claude-opus-4-20250514200K
Claude Sonnet 4claude-sonnet-4-20250514200K

The [1m] twins. /v1/models lists every 1M-context model a second time with a [1m] suffix, for example claude-opus-5-5[1m]. That suffix is for Claude Code, which sizes its context budget from a built-in model table rather than from this list. GrowMoon strips it before routing, so it changes neither the model nor its context window.

Aliases. The shorthands opus, sonnet and haiku also resolve, which is handy for tools that only accept a family name.

Troubleshooting

Connection errors

Check that the key is active and has not expired. The Check key page tells you in one look.

Web search or image tools not responding

They run server-side, so there is nothing local to fix. Confirm the key is valid and retry.

Model not found

Use an exact ID from the models list above. Display names are not accepted, only IDs and the three family aliases.

Rate limited

The 5-hour token window may be used up. Check the key to see how long is left before it resets.

Changes not applying

Restart the editor after any configuration change. Most read their settings only at launch.

Cursor or Windsurf not routing

Those two need the /v1 endpoint URL, not the bare host.

Still stuck

Keys are issued and managed by GrowMoon and its authorised resellers. Whoever gave you your key can see its usage, reset its window, or issue a replacement.