Developer API

Overview

Connect Claude Code CLI, Claude Desktop, Cursor, VS Code extensions, Excel, and any OpenAI or Anthropic compatible tool to your plan. Requests run on the same high-speed models and daily token allowance as your dashboard.

1. Create a key

1. Sign in to the ChatGPT Plus or Claude Pro dashboard for the subscription you want to use, then choose API keys in the sidebar.
2. Give the key a name (for example "Claude Code CLI" or "Cursor Laptop") and select Create key.
3. Copy the key right away. It starts with sk-cp- and is shown only once.

The Connect tab in the same window has copy-ready settings for every tool on this page. Each dashboard shows only its own subscription's keys. Revoke a key there at any time; it stops working on the next request. When a subscription expires, its existing keys pause and the same keys resume after renewal.

2. Models

Your key lists only the models included in your plan. Call GET /models to see them. Model IDs are case-insensitive, and official model names or aliases work seamlessly.

ChatGPT Plus Models

Claude Pro Models

3. Limits & Real Token Counter

• Token Accounting: Each request deducts prompt and completion tokens from your daily allowance. Provider-reported counts are used when available; otherwise the gateway reports an estimate.
• Max Output Tokens: Supports up to 128,000 output tokens per response (max_tokens / max_completion_tokens) matching ChatGPT Plus & Claude Pro flagship specifications, with unbuffered HTTP streaming.
• Maximum Context Window: Up to 1,050,000 tokens context window for massive prompts, long codebases, and deep document reasoning.
• Seamless Multi-Part Auto-Continuation: Prompts needing massive outputs automatically chain up to 5 segments continuously without cuts, truncation, or broken responses.
• Autonomous Live Web Search: Real-time web queries and location search trigger automatically when fresh information, news, or live data is requested.
• Key Limits: The default key allowance is 1,000,000 tokens per day or your larger assigned daily plan allowance, with 240 requests per minute. Admin settings may change these values.
• Keys per subscription: Up to 10 active keys by default for each plan. Each dashboard tracks that subscription's token consumption.
• Daily counters reset at 00:00 UTC. A limited request returns HTTP 429 with Retry-After.

4. Quick test

List your models, then send a message via OpenAI or Anthropic format. Replace YOUR_API_KEY with your key.

curl {{BASE}}/models \
  -H "Authorization: Bearer YOUR_API_KEY"

OpenAI Chat Completions:

curl {{BASE}}/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","messages":[{"role":"user","content":"Hello"}]}'

Anthropic Messages (Claude):

curl {{BASE}}/messages \
  -H "x-api-key: YOUR_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-3-7-sonnet","max_tokens":1024,"messages":[{"role":"user","content":"Hello from Claude"}]}'

5. Claude Code CLI (Agent & Tool Calling)

Anthropic's official Claude Code CLI agent works directly with your Auvryn Claude Pro key. All agent capabilities—including automatic file reading, editing, diffing, bash execution, and project analysis—are fully supported via our native Anthropic SSE streaming tool engine.

1. Install Claude Code globally via npm:

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

   Copy
2. Configure your environment variables with your Auvryn Claude Pro API key:

   Linux / macOS (Bash / Zsh):

   export ANTHROPIC_BASE_URL="{{BASE}}"
   export ANTHROPIC_API_KEY="YOUR_API_KEY"

   Copy

   Windows (PowerShell):

   $env:ANTHROPIC_BASE_URL = "{{BASE}}"
   $env:ANTHROPIC_API_KEY = "YOUR_API_KEY"

   Copy

   Windows (CMD):

   set ANTHROPIC_BASE_URL={{BASE}}
   set ANTHROPIC_API_KEY=YOUR_API_KEY

   Copy
3. Launch Claude Code in your project directory:

   claude

   Copy

Autonomous Mode & Permission Settings (Zero Popups)

By default, Claude Code pauses and asks for permission on every file change or terminal command. To enable 100% autonomous coding with zero interruptions:

Option A: Fast Autonomous Launch (Bypass Permissions)

Launch Claude with the skip flag, or press Shift + Tab inside Claude Code to switch permission mode to Bypass permissions:

claude --dangerously-skip-permissions

Option B: Permanent Auto-Approve via settings.json (Recommended)

Save this in your user config ~/.claude/settings.json (or project-level .claude/settings.json):

{
  "permissions": {
    "allow": [
      "Bash(*)",
      "FileEdit(*)",
      "FileWrite(*)",
      "Read(*)",
      "Glob(*)",
      "Grep(*)",
      "*"
    ]
  }
}

Resolving "Locked by Organization Policy" & Package-Manager Network Access (npm & pip)

If you or your tools encounter a message saying "The Local sandbox and Strict sandbox mode toggles are locked by organization policy" when installing project dependencies (e.g. for video rendering, FFmpeg, Remotion, MoviePy, or Python science packages), here is why it occurs and how to unlock it in 30 seconds:

Method 1: Run with Autonomous Bypass (Fastest Fix)

Launch Claude Code with the bypass flag. This immediately removes local sandbox blocks and gives Claude Code full network access to install npm and pip packages:

claude --dangerously-skip-permissions

Shortcut: If Claude Code is already running, press Shift + Tab in your terminal to switch permission mode to Bypass permissions.

Method 2: Pre-Authorize npm & PyPI in settings.json (Permanent Fix)

To whitelist package-manager egress permanently so Claude Code can install any dependency without popups, save this in your configuration file:
• Windows: %USERPROFILE%\.claude\settings.json (or C:\Users\<Username>\.claude\settings.json)
• macOS / Linux: ~/.claude/settings.json

{
  "permissions": {
    "allow": [
      "Bash(npm *)",
      "Bash(npx *)",
      "Bash(pnpm *)",
      "Bash(yarn *)",
      "Bash(pip *)",
      "Bash(pip3 *)",
      "Bash(python *)",
      "Bash(python3 *)",
      "Bash(ffmpeg *)",
      "Bash(*)",
      "FileEdit(*)",
      "FileWrite(*)",
      "Read(*)",
      "Glob(*)",
      "Grep(*)",
      "*"
    ]
  }
}

For Video-Rendering & Motion Graphics Projects (Remotion, FFmpeg, MoviePy):

Video pipelines frequently compile binaries and pull heavy media packages. You can verify network egress by having Claude install standard video libraries:

# Node.js Video Rendering (Remotion / Canvas)
npm install --save-dev @remotion/cli @remotion/renderer react react-dom

# Python Video Automation (MoviePy / FFmpeg / OpenCV)
pip install moviepy opencv-python pillow numpy

Checking License Validity & Plan Expiration

You can verify your subscription validity, active days remaining, and daily token quota at any time using any of these methods:

• Directly inside Claude Software: In Claude Code CLI or Claude Desktop, ask Claude: "What is my plan validity?" or type /status. Claude will directly output your verified plan name, expiry date, and days remaining.
• Live Browser Certificate: Open https://your-domain/v1/status?key=YOUR_API_KEY in your browser for an official verified green badge and expiration certificate.
• Dashboard API Keys Modal: Click API Keys in your dashboard to view your live active subscription banner and days remaining.

6. Claude Desktop App (PC / Laptop / Mac)

Connect the official Claude Desktop app directly to your Auvryn subscription using either Developer Mode or configuration file:

Method A: App Developer Mode (Recommended)

1. Open the Claude app on your PC, Laptop, or Mac.
2. Click Help → Troubleshooting.
3. Enable Developer Mode.
4. Click Developer Mode in the menu.
5. Select Configure Third-Party Configuration.
6. Credential Kind/Type: Static API Key.
7. Gateway / Base URL: https://your-domain/api/v1
8. API Key: Paste your generated Auvryn API key.
9. Header / Auth: Select x-api-key.
10. Click Save Changes → Click Apply Changes.

Method B: Configuration File (claude_desktop_config.json)

Config file path:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "C:\\Users\\your-username\\Desktop"]
    },
    "puppeteer": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-puppeteer"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "YOUR_GITHUB_TOKEN" }
    }
  },
  "env": {
    "ANTHROPIC_BASE_URL": "{{BASE}}",
    "ANTHROPIC_API_KEY": "YOUR_API_KEY"
  }
}

7. Cursor, OpenAI Codex & Desktop AI Apps

Use your plan inside Cursor IDE, OpenAI Codex desktop software, Chatbox, Jan AI, NextChat, or any OpenAI-compatible client on PC/Laptop/Mac.

Desktop Software (Chatbox, Jan AI, NextChat)

1. Open your ChatGPT / Codex desktop application.
2. Open Settings ⚙️ → Model Provider / AI Settings.
3. Select Provider: OpenAI API Compatible.
4. Credential Type: Bearer API Key.
5. Gateway / Base URL: https://your-domain/api/v1
6. API Key: Paste your Auvryn API key.
7. Auth Header: Select Authorization: Bearer.
8. Model ID: gpt-4o (or sol, luna, astra).
9. Click Save Changes → Apply / Connect.

Cursor IDE Configuration

1. In Cursor, open Settings (Gear icon or Ctrl+, / Cmd+,) and navigate to Models.
2. Turn on OpenAI API Key or Custom OpenAI Base URL.
3. Set Override OpenAI Base URL to: https://your-domain/api/v1
4. Paste your API key starting with sk-cp- into the API Key field.
5. Under Model Names, add the models you wish to use: gpt-6, astra, gpt-4o, sol, gpt-4o-mini, or claude-3-7-sonnet.

8. VS Code: Continue (Agent Mode & Tools)

1. Install Continue from the VS Code Extensions view.
2. Open the Continue panel, open the model selector and choose to add a model through the config file, or open ~/.continue/config.yaml (on Windows: %USERPROFILE%\.continue\config.yaml) directly.
3. Add a model entry with capabilities: [tool_use] (and image_input for vision models) like the one below, save, and pick it in Continue. You can then use Continue in Agent mode for automated file reads, edits, diffs, and local terminal execution.

name: Developer API
version: 1.0.0
schema: v1
models:
  - name: GPT-6 Astra
    provider: openai
    model: astra
    apiBase: {{BASE}}
    apiKey: YOUR_API_KEY
    contextLength: 1000000
    capabilities:
      - tool_use
      - image_input
    roles:
      - chat
      - edit
      - apply
  - name: Claude 3.7 Sonnet
    provider: openai
    model: claude-3-7-sonnet
    apiBase: {{BASE}}
    apiKey: YOUR_API_KEY
    contextLength: 1048576
    capabilities:
      - tool_use
      - image_input
    roles:
      - chat
      - edit
      - apply

Older Continue versions use config.json: add {"title": "GPT-4o", "provider": "openai", "model": "gpt-4o", "apiBase": "…/api/v1", "apiKey": "…"} to models.

9. VS Code: Cline, Roo Code and Kilo Code

All three extensions use the same "OpenAI Compatible" provider.

1. Install the extension and open its settings (the gear icon in its panel).
2. Set API Provider to OpenAI Compatible.
3. Base URL: https://your-domain/api/v1
4. API Key: your sk-cp-… key.
5. Model ID: a model from your plan, for example astra, sol, or sonnet.
6. Optional model settings: max output tokens 8192 or less; turn off image support for models without images; a context window of 128000 is a safe setting.

10. Excel add-in

The add-in opens a task pane where you paste your API key once, pick a model and then:

Manifest

Install (sideload)

Excel on the web

1. Download manifest.xml to your computer.
2. Open a workbook at office.com, then go to Home › Add-ins › More Add-ins (or Insert › Add-ins).
3. Select My Add-ins › Upload My Add-in, choose the downloaded file and select Upload.

Excel for Windows (Microsoft 365)

1. Create a folder, share it on your network (for example \\YOUR-PC\addins) and save manifest.xml in it.
2. In Excel open File › Options › Trust Center › Trust Center Settings › Trusted Add-in Catalogs.
3. Enter the share path as the Catalog Url, select Add catalog, tick Show in Menu, then restart Excel.
4. Go to Home › Add-ins › More Add-ins › Shared Folder, choose the add-in and select Add.

Excel for Mac

1. Copy manifest.xml to ~/Library/Containers/com.microsoft.Excel/Data/Documents/wef (create the wef folder if needed).
2. Restart Excel, then open Home › Add-ins (or Insert › My Add-ins) and choose the add-in.

Whole organization: a Microsoft 365 admin can deploy it from Admin center › Settings › Integrated apps › Upload custom apps using the manifest URL above.

Works in Excel for Microsoft 365 or Excel 2021 and later on Windows, Excel for Mac and Excel on the web. Older Excel versions that use the Internet Explorer engine for add-ins are not supported.

Use it

1. Open the add-in from the Home tab (Open assistant) or from My Add-ins.
2. Paste your API key and select Save key. It's stored only in this Excel profile on this device.
3. Choose a model, select some cells, type a question or instruction and pick an action.

11. Errors

Errors follow standard OpenAI and Anthropic JSON error formats: {"error": {"message", "type", "param", "code"}}.

If a stream has already started, an error arrives as a final data: {"error": …} event followed by data: [DONE].

12. Compatibility notes

• Supported endpoints: /v1/chat/completions (OpenAI format), /v1/messages (Anthropic format), /v1/models.
• Agent Tool Calling: Full support for native OpenAI function calling (tools, tool_choice, parallel_tool_calls) and Anthropic tool use (tools, tool_use blocks, input JSON streaming). Perfect for Claude Code CLI, Cursor Agent, Continue Agent, and Cline/Roo Code.
• Streaming: Server-sent events (SSE) with unbuffered chunking (X-Accel-Buffering: no) and stream_options.include_usage support.
• Output Tokens: max_tokens / max_completion_tokens can request up to 128,000 tokens; actual output also depends on the selected provider's capacity.
• Roles: system, developer, user, assistant, and tool execution results.
• Multimodal Images: Supported across all models via base64 data URLs (data:image/png;base64,…) and Anthropic image blocks.
• Usage Tracking: Prompt and completion usage is reported from upstream when available, otherwise estimated by the gateway.