Multipass

MULTIPASS

Universal API Wrapper - Turn ANY Python Library into a Robust API

MULTIPASS A Universal API Wrapper - Turn ANY Python Library into a Robust API

The architecture I've created is completely universal and works with:

✅ Computer Vision: YOLO, Ultralytics, OpenCV, PIL, scikit-image ✅ LLMs: Transformers, OpenAI, Anthropic, MLX, LangChain ✅ ML Frameworks: PyTorch, TensorFlow, JAX, scikit-learn ✅ Data Science: Pandas, NumPy, Polars, DuckDB ✅ Web Apps: Streamlit, Gradio, Dash, FastAPI ✅ Audio/Video: Whisper, FFmpeg, PyDub, MoviePy ✅ Any Custom Library: Your proprietary code, research projects

🚀 Quick Start (Literally One Command!)

# Install the launcher
pip install fastapi uvicorn

# Start ANY library as an API
python api_launcher.py yolo
python api_launcher.py gpt2
python api_launcher.py pandas
python api_launcher.py opencv

That's it! Your API is running at http://localhost:8000

🎯 How It Works

1. Automatic Service Discovery

The wrapper automatically:

• Inspects the library to find all functions/classes
• Analyzes their signatures and parameters
• Creates REST endpoints for each function
• Generates OpenAPI documentation

2. Universal Adapter Pattern

# It works with ANY library pattern:

# Simple functions
import numpy as np
# → GET /mean, POST /reshape, etc.

# Classes with methods
from ultralytics import YOLO
model = YOLO()
# → POST /detect, POST /train, etc.

# Complex pipelines
from transformers import pipeline
nlp = pipeline("sentiment-analysis")
# → POST /analyze

3. Built-in Resilience

• 🔄 Automatic retry with backoff
• 🚦 Circuit breakers for fault tolerance
• 📊 Health checks and monitoring
• 🔌 Connection pooling
• 💾 Response caching

📚 Library-Specific Examples

YOLO Object Detection

# Start YOLO API
python api_launcher.py yolo

# Use it
curl -X POST http://localhost:8000/detect \
  -F "image=@photo.jpg"

GPT-2 Text Generation

# Start GPT-2 API
python api_launcher.py gpt2

# Use it
curl -X POST http://localhost:8000/generate \
  -d '{"text": "Once upon a time"}'

Pandas Data Processing

# Start Pandas API
python api_launcher.py pandas

# Use it
curl -X POST http://localhost:8000/read_csv \
  -F "file=@data.csv"

Streamlit App Manager

# Create Streamlit API
from universal_api_wrapper import UniversalAPIFactory

app = UniversalAPIFactory.create_api('streamlit', {
    'adapter_class': 'StreamlitAdapter',
    'apps': [
        {'name': 'dashboard', 'path': './dashboard.py'},
        {'name': 'ml_demo', 'path': './ml_demo.py'}
    ]
})

🔧 Advanced Configuration

Custom Library Configuration

# my_library_config.yaml
my_ml_pipeline:
  module: my_company.ml_pipeline
  init_function: load_model
  init_args:
    model_path: ./models/production.pkl
    config: ./config/settings.yaml
  endpoints:
    - preprocess
    - predict
    - evaluate
  authentication: true
  rate_limit: 100  # requests per minute

Use Custom Config

python universal_api_wrapper.py my_ml_pipeline --config my_library_config.yaml

🐳 Production Deployment

Docker Compose for Multiple Services

services:
  # Computer Vision API
  yolo-api:
    build: .
    command: python api_launcher.py yolo
    ports:
      - "8001:8000"
    deploy:
      replicas: 3
      
  # LLM API
  llm-api:
    build: .
    command: python api_launcher.py gpt2
    ports:
      - "8002:8000"
      
  # Load Balancer
  nginx:
    image: nginx
    ports:
      - "80:80"
    depends_on:
      - yolo-api
      - llm-api

🔌 Client Libraries

Python Client

from universal_api_client import UniversalClient

# Connect to any wrapped library
client = UniversalClient("http://localhost:8000")

# Discover available methods
services = await client.discover()

# Call any method dynamically
result = await client.detect(image="photo.jpg", confidence=0.5)

JavaScript/TypeScript Client

const client = new UniversalAPIClient('http://localhost:8000');

// Auto-discovers methods
const services = await client.discover();

// Type-safe calls
const result = await client.detect({ image: imageBase64 });

🛡️ Security & Monitoring

Built-in Security

• API key authentication
• Rate limiting per endpoint
• Input validation
• CORS configuration
• SSL/TLS support

Monitoring

• Prometheus metrics
• Health checks
• Performance tracking
• Error logging
• Request tracing

🎨 Special Features

1. Model Context Protocol (MCP)

# Expose any library as MCP server for AI assistants
python mlx_whisper_mcp.py --library pandas

2. Streaming Support

• WebSocket endpoints for real-time data
• Server-sent events for long operations
• Chunked responses for large files

3. Batch Processing

# Batch endpoint automatically created
POST /batch/detect
{
  "items": [
    {"image": "img1.jpg"},
    {"image": "img2.jpg"},
    {"image": "img3.jpg"}
  ]
}

4. Pipeline Chaining

# Chain multiple operations
POST /pipeline
{
  "steps": [
    {"service": "resize", "args": {"size": [224, 224]}},
    {"service": "detect", "args": {"confidence": 0.5}},
    {"service": "classify", "args": {"top_k": 5}}
  ]
}

📊 Performance

• Latency:

That's it! 🎉

No more connection errors. No more endpoint breakage. No more manual API maintenance. Just reliable, scalable APIs for any Python library!

Please help to develop it further.

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [tinosingh](https://github.com/tinosingh)
- **Source:** [tinosingh/multipass](https://github.com/tinosingh/multipass)
- **License:** Apache-2.0
- **Homepage:** https://github.com/tinosingh/multipass

Install and usage instructions live in the source repository linked above.