novai-python
# novai-python
<!-- mcp-name: io.github.vvvvking/novai-python -->
OpenAI-compatible **SDK + MCP server** for 120+ Chinese frontier AI models
(DeepSeek, Qwen, GLM, Kimi, Seedance, MiniMax, Hunyuan, CogVideoX, ...) via
[NovAI](https://aiapi-pro.com) — one key, one endpoint, **0% platform fee**
(raw provider list price; vs OpenRouter ~5.5% markup).
- Endpoint: `https://aiapi-pro.com/v1` (change one `base_url` line from any OpenAI code)
- Free: `$2` signup credit, no card; plus 5 free models (no credit charge; 10 calls/day on the free tier)
(`glm-4.7-flash`, `glm-4.6v-flash`, `glm-4.1v-thinking-flash`, `cogview-3-flash`, `cogvideox-flash`)
- Live pricing: <https://aiapi-pro.com/china-ai-leaderboard.json>
## Install
PyPI package name: `aiapi-pro` (import name: `novai`). Once published:
```bash
pip install aiapi-pro
export NOVAI_API_KEY=nvai-... # https://aiapi-pro.com/register ($2 free)
```
From source (always current):
```bash
pip install git+https://github.com/vvvvking/novai-python.git
export NOVAI_API_KEY=nvai-... # https://aiapi-pro.com/register ($2 free)
```
## SDK quickstart
```python
from novai import chat, list_models, generate_image, pricing
print(chat("deepseek-v4-flash", "Explain MCP in one sentence."))
print(generate_image("cogview-3-flash", "a red panda astronaut")) # free model
print(pricing()["categories"][0]["models"][0]) # cheapest chat model, live
```
Any existing OpenAI code works by swapping the base URL:
```python
from openai import OpenAI
c = OpenAI(base_url="https://aiapi-pro.com/v1", api_key="nvai-...")
```
## MCP server
Expose the models as tools to any MCP client (Claude Desktop, Cursor, Cline, ...):
```json
{
"mcpServers": {
"novai": {
"command": "python",
"args": ["-m", "novai.mcp_server"],
"env": { "NOVAI_API_KEY": "nvai-..." }
}
}
}
```
Tools provided: `novai_list_models`, `novai_chat`, `novai_generate_image`,
`novai_free_models`, `novai_cheapest`, `novai_pricing_feed`, `novai_transparency`.
## Transparency
0% platform fee, 1.0x price multiplier, $5 min top-up, 60 req/min, unused-balance
7-day refund, failed video/3D jobs refunded in full. Machine-readable:
<https://aiapi-pro.com/transparency.json>. Chinese image models apply a visible
"AI-generated" corner watermark per China labeling rules
(<https://aiapi-pro.com/watermark-notice>).
## License
MIT
TDQS
Scored across 7 tools
Each tool has a distinct purpose, and the descriptions explicitly resolve the near-overlaps: novai_list_models vs novai_free_models, and novai_cheapest vs novai_pricing_feed each point to the other with clear 'use X instead' guidance. Chat and image generation are cleanly separated by modality.
All tools share the novai_ prefix and snake_case, which is highly predictable. Minor deviation: some names are verb-led (novai_chat, novai_list_models, novai_generate_image) while others are noun-led (novai_free_models, novai_cheapest, novai_pricing_feed, novai_transparency), but each is still self-explanatory.
Seven tools is well within the ideal range and each one earns its place: model discovery, chat, image generation, and three distinct pricing/policy views. No redundancy or filler.
Core workflows (discover models, chat, generate images, check pricing/policy) are covered, but the pricing tool exposes Video, 3D, Audio, and Embedding categories with no corresponding callable tools, and there is no multi-turn/conversation-state or token-accounting support. These are notable gaps an agent would hit when users ask for non-chat modalities.