Connect TokenPAPA to AI Agents

You already use TokenPAPA for one simple reason: one API key, every major model — DeepSeek, Qwen, MiniMax, Kimi, Claude and more, with no Chinese phone number and no geo-restrictions.

But a key in your dashboard is useless until it's inside the tools you actually work with. And these days, that means agents — Cursor auto-completing your code, Claude Code refactoring your repo, Codex turning a TODO into a pull request.

Here's the good news: because TokenPAPA speaks the OpenAI-compatible protocol, almost every agent on the market can connect to it in under two minutes. No SDK rewrites, no plugins, no vendor lock-in.

This guide walks you through the universal setup pattern, then gives you copy-paste configurations for the most popular coding agents and AI tools of 2026. Every section ends with a success check so you always know whether you got it right.

What You Need

Before you start, grab these three things from your TokenPAPA dashboard:

⚠️ Save your key when you create it. For security reasons, TokenPAPA only displays a new key once. If you lose it, revoke and create a new one.

The Universal Pattern: 3 Fields, Any Tool

Every OpenAI-compatible agent needs the same three fields. Once you understand this pattern, you can configure any tool without a guide:

Base URL  →  https://tokenpapa.ai/v1
API Key   →  sk-xxx...xxxx
Model     →  deepseek-v4-flash   (or any model in your plan)

That's it. Whether the tool is a coding agent, a chat client, or an automation platform, it's always these three fields. If an app asks for "OpenAI API key", "custom provider", or "API-compatible endpoint" — that's your cue.

Step 0: Verify Your Key Before Touching Any GUI

Before configuring anything, run this one command in your terminal. It confirms that your Key, Base URL and model name are all correct — so when a tool doesn't work later, you'll know the problem is in the tool's settings, not your credentials:

curl https://tokenpapa.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-xxx...xxxx" \
  -d '{"model": "deepseek-v4-flash", "messages": [{"role": "user", "content": "ping"}]}'

Success check: you get a JSON response back containing the model's reply. If you get 401 Unauthorized → your key is wrong or expired. If you get 404 Model Not Found → the model ID is wrong — copy the exact ID from Model Square.

Once this passes, every tool below is just "paste these three values into different boxes."

Coding Agents

Cursor

Cursor is one of the most popular AI code editors, and it supports custom OpenAI-compatible API providers.

Setup:

1. Open Cursor → Settings (gear icon) → Models
2. Scroll to the OpenAI API Key section → toggle Override OpenAI Base URL
3. Enter:
   • OpenAI API Key: sk-xxx...xxxx
   • OpenAI Base URL: https://tokenpapa.ai/v1
4. Click Verify — Cursor will confirm the connection
5. Add a model name (e.g. deepseek-v4-flash) to your model list, then select it in the model picker

✅ Success check: the Verify button shows a green confirmation, your model appears in the model picker, and a chat message gets a reply.

💡 Can't find the option? In Cursor's Settings, use the search box and type openai base url to jump straight to the field. Its exact position varies slightly between versions. 💡 Prefer environment variables? Set OPENAI_API_KEY and OPENAI_BASE_URL before launching Cursor from your terminal — many teams do this to keep keys out of the UI.

Claude Code

Claude Code (Anthropic's terminal agent) normally talks to Anthropic's API. If your TokenPAPA plan includes Claude models and exposes an Anthropic-compatible endpoint, point Claude Code at TokenPAPA with two environment variables:

export ANTHROPIC_BASE_URL="https://tokenpapa.ai"   # ← your Anthropic-compatible endpoint (no /v1 — the client appends it)
export ANTHROPIC_AUTH_TOKEN="sk-xxx...xxxx"
export ANTHROPIC_MODEL="claude-sonnet-4-6"             # or whichever Claude model you use
claude

✅ Success check: Claude Code starts normally and answers a test prompt with no 401 errors.

⚠️ Check before you run: confirm the Anthropic-compatible endpoint and Claude model IDs in your TokenPAPA model list / API docs. If your account uses TokenPAPA purely via the OpenAI-compatible endpoint, use Cursor, Codex CLI or Cline (below) with Claude models instead — same models, no protocol mismatch.

OpenAI Codex CLI

OpenAI's open-source Codex CLI agent respects standard OpenAI environment variables, so pointing it at TokenPAPA is trivial:

export OPENAI_API_KEY="sk-xxx...xxxx"
export OPENAI_BASE_URL="https://tokenpapa.ai/v1"

# run with your model of choice
codex exec --model deepseek-v4-flash "explain this repo"

✅ Success check: Codex responds to your prompt normally, with no 401/404 errors in the output.

💡 TokenPAPA carries DeepSeek, Qwen, Kimi and MiniMax models at a fraction of GPT pricing — using them inside Codex is one of the cheapest ways to run a terminal agent in 2026. See the LLM API Pricing Comparison for numbers.

Cline & Roo Code (VS Code)

Cline and Roo Code are the most popular open-source AI coding agents for VS Code. Both support any OpenAI-compatible provider.

Setup (same for both):

1. Install the extension → open its settings panel
2. API Provider: select OpenAI Compatible
3. Fill in:
   • Base URL: https://tokenpapa.ai/v1
   • API Key: sk-xxx...xxxx
   • Model ID: deepseek-v4-flash (or qwen3.7-max for agentic coding)
4. Click the "Check" button to verify — you should see the model's pricing appear

✅ Success check: the Check button returns a green ✓ (or shows the model's price). If the button is labelled "Connect" instead of "Check" on your version, that's fine — the green confirmation is what matters.

Continue (VS Code / JetBrains)

Continue is a popular open-source AI assistant that reads a config file for providers.

Add this to your config.json:

{
  "models": [
    {
      "title": "DeepSeek V4 Flash (TokenPAPA)",
      "provider": "openai",
      "model": "deepseek-v4-flash",
      "apiBase": "https://tokenpapa.ai/v1",
      "apiKey": "sk-xxx...xxxx"
    }
  ]
}

✅ Success check: after restarting the extension, "DeepSeek V4 Flash (TokenPAPA)" appears in the model dropdown and answers a test prompt.

Workflow & No-Code Agents

Dify

Dify lets you build AI apps visually, and any OpenAI-compatible model provider plugs straight in.

1. Go to Settings → Model Provider
2. Find OpenAI-API-compatible (or add a custom provider)
3. Enter:
   • API Base URL: https://tokenpapa.ai/v1
   • API Key: sk-xxx...xxxx
4. Click Save — Dify fetches your available models automatically

✅ Success check: after saving, your available TokenPAPA models appear in the model list, and you can select them inside any workflow or chatbot app.

💡 Can't find "OpenAI-API-compatible" in the list? Pick any provider that says "OpenAI-compatible" — the fields are identical.

n8n

n8n is the leading open-source workflow automation platform. Use TokenPAPA inside its OpenAI node:

1. Create an OpenAI node credential
2. Click "Use custom base URL"
3. Set:
   • Base URL: https://tokenpapa.ai/v1
   • API Key: sk-xxx...xxxx
4. In the node, set Model to e.g. deepseek-v4-flash

✅ Success check: running a test execution of the node returns a successful result (green), not an auth error.

💡 Tip: many n8n nodes (LangChain, AI Agent, Basic LLM Chain) are OpenAI-compatible under the hood and expose a custom base URL field — the same three fields work everywhere.

Chat Apps & One-Click Import

For chat clients (LobeChat, OpenCat, Cherry Studio, ChatBox, Open WebUI…), you don't even need this guide: TokenPAPA's Console → Token page has a one-click Import button for supported apps, and every OpenAI-compatible chat client accepts the same Base URL + Key + Model trio.

See the Chat Application Integration guide for the full list and manual steps.

Which Model Should Each Agent Use?

A quick cheat sheet based on what each tool does best:

Your exact model list depends on your plan — check Model Square on the Pricing page for the current catalog and per-model pricing.

Troubleshooting

If a tool fails after setup, work through this table top to bottom:

Golden rule: if Step 0's curl test passes but a tool still fails, the problem is in that tool's settings — re-check the three fields you entered there. For deeper API-level debugging, see LLM API Error Handling & Debugging Guide.

Security Reminders

• Never commit keys to git. Use environment variables or the tool's secret store
• One key per tool where possible — TokenPAPA lets you create multiple tokens, so revoke a single leaked key without touching the rest
• Rotate regularly. Create a new key → update your tools → delete the old one
• Set spend awareness: check usage in Console → Logs to catch unexpected consumption early

Start Building

You now have one key powering your entire AI toolkit — the cheapest Chinese LLMs (DeepSeek V4 Flash at $0.14/1M input) inside the same agents people pay 10–20x more for elsewhere.

Get started in 2 minutes:

1. Sign up for TokenPAPA
2. Create your API key in the Console
3. Run the Step 0 curl test, then pick an agent above and paste the three fields

Questions? Drop by our GitHub or reach support via the dashboard — happy building! 🚀