MiniMax2API

OpenAI-compatible API proxy for the MiniMax Agent API.

English · Bahasa Indonesia · 中文 · 日本語 · Español

---

What is this?

MiniMax2API turns the MiniMax Agent web API (agent.minimax.io) into a standard OpenAI-compatible endpoint, so you can use it with any OpenAI SDK, client, or router (including 9router, OpenWebUI, LibreChat, LangChain, and more).

It handles the reverse-engineered request signing, session management, and Server-Sent-Events streaming for you — you just talk OpenAI.

Built on the same idea as Duck2API: a small, self-hosted, drop-in OpenAI proxy.

Features

Feature	Endpoint	Status
Chat Completions	POST /v1/chat/completions	✅
Streaming (SSE)	POST /v1/chat/completions (stream: true)	✅
Model listing	GET /v1/models	✅
Health check	GET /health	✅
Bearer / x-api-key auth	all routes	✅
Multi-turn context	flattened into the prompt	✅
Docker / Compose	included	✅

Quick start

1. Get your credentials

Log in at https://agent.minimax.io/, open DevTools → Console, and run:

python app/get_credentials.py --json

or paste the snippet from python app/get_credentials.py directly into the console. It prints something like:

{"token":"eyJhbGci...","agent_id":"447316105384064","uuid":"1604265c-...","device_id":"52668348","user_id":"561636901788553221"}

2. Configure

cp env.template .env
# then edit .env and fill in MM_TOKEN, MM_AGENT_ID, MM_UUID, MM_DEVICE_ID, MM_USER_ID

3. Run

Local:

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cd app && python server.py

Docker:

docker build -t minimax2api .
docker run -d --name minimax2api --env-file .env -p 8100:8100 minimax2api

Docker Compose:

docker compose up -d

Usage

cURL

curl http://localhost:8100/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-minimax-local" \
  -d '{
    "model": "MiniMax-M3.1-Flash-Preview",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

OpenAI Python SDK

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8100/v1", api_key="sk-minimax-local")

stream = client.chat.completions.create(
    model="MiniMax-M3.1-Flash-Preview",
    messages=[{"role": "user", "content": "Write a haiku about proxies."}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Full endpoint reference: API.md.

How it works

OpenAI client ──▶ FastAPI (/v1/chat/completions)
                      │
                      ▼
             MiniMaxAdapter  (app/adapter.py)
                      │  signs requests (x-signature, yy)
                      ▼
   POST /agent/{id}/session      → create session
   POST /session/{id}/message    → SSE stream of the reply

The MiniMax web app signs every request with a timestamped x-signature and a yy fingerprint. app/sign.py reproduces that algorithm; app/adapter.py drives the session/turn lifecycle and parses the reply stream.

Configuration

See env.template for every option. The important ones:

Variable	Required	Description
MM_TOKEN	✅	MiniMax JWT
MM_AGENT_ID	✅	Agent id from the /mavis?id= URL
MM_UUID	✅	localStorage.UNIQUE_USER_ID
MM_DEVICE_ID	✅	sessionStorage.tab_device_id
MM_USER_ID	✅	localStorage.user_detail_agent.realUserID
MM_MODEL	–	Model id (default MiniMax-M3.1-Flash-Preview)
MM_API_KEY	–	Require this key on requests (empty = no auth)
PORT	–	Listen port (default 8100)

Project layout

.
├── app/
│   ├── adapter.py         # MiniMax session + streaming logic
│   ├── server.py          # FastAPI OpenAI-compatible server
│   ├── sign.py            # request signing (x-signature, yy)
│   ├── get_credentials.py # helper to extract browser credentials
│   ├── tool_dsml.py       # tool-call parsing helpers
│   └── tool_sieve.py
├── assets/                # logo & banner
├── API.md                 # full API reference
├── Dockerfile
├── docker-compose.yml
├── env.template
└── requirements.txt

Contributing

Issues and pull requests are welcome. Please keep changes focused and test the server against a live MiniMax session before opening a PR.

Disclaimer

This project is unofficial and not affiliated with MiniMax. It is intended for personal and educational use. Respect MiniMax's Terms of Service. Your credentials stay on your machine — this server only talks to agent.minimax.io on your behalf.

License

MIT