Flux Protocol

Universal AI Connectivity Layer - Connect AI to anything in 4 lines. 40% fewer tokens with TOON format.

npm License: MIT

Why FLUX?

Feature	MCP	FLUX
Lines of code	120+	4
Protocols	STDIO only	STDIO, HTTP, WebSocket
Token efficiency	JSON	TOON (40% fewer tokens)
Type safety	Manual	Automatic

Install

npm install fluxprotocol-sdk

Quick Start

npx fluxprotocol-cli create weather
cd weather && pnpm install && pnpm dev

The 4-Line Promise

import { connector, method, FluxServer, StdioTransport } from 'fluxprotocol-sdk';

@connector('weather')
class WeatherConnector {
  @method()
  async getWeather(city: string) {
    return fetch(`https://api.weather.com/${city}`).then(r => r.json());
  }
}

const server = new FluxServer(WeatherConnector);
server.setTransport(new StdioTransport(server));
server.start();

Packages

Package	Description	Install
fluxprotocol-sdk	Core SDK with transports	npm i fluxprotocol-sdk
fluxprotocol-cli	CLI tools	npm i -g fluxprotocol-cli
fluxprotocol-client	HTTP/WS client	npm i fluxprotocol-client

Transports

import { StdioTransport, HttpTransport, WebSocketTransport } from 'fluxprotocol-sdk';

new StdioTransport(server);                     // MCP (Claude Desktop)
new HttpTransport(server, { port: 3000 });      // REST API + TOON
new WebSocketTransport(server, { port: 3001 }); // Real-time

TOON Format (40% Fewer Tokens)

FLUX supports TOON (Token-Oriented Object Notation) for efficient AI communication:

import { FluxClient } from 'fluxprotocol-client';

const client = new FluxClient('http://localhost:3000');

// Enable TOON format (40% fewer tokens)
const tools = await client.useToon().listTools();

// Or use JSON (default)
const weather = await client.call('weather.getWeather', { city: 'Tokyo' });

Server auto-detects format via Accept header:

• Accept: application/toon → TOON response
• Accept: application/json → JSON response (default)

Production Decorators

import { cache, rateLimit, retry, config } from 'fluxprotocol-sdk';

@connector('api')
@config({ auth: { type: 'bearer', env: 'API_TOKEN' } })
class ApiConnector {
  @method()
  @cache({ ttl: 60000 })                          // Cache 1 minute
  @rateLimit({ requests: 100, window: '1m' })     // 100 req/min
  @retry({ attempts: 3, backoff: 'exponential' }) // Auto-retry
  async fetchData(id: string) { ... }
}

Decorator	Purpose
@cache	LRU cache with TTL
@rateLimit	Token bucket rate limiting
@retry	Exponential/linear backoff
@config	Auth & environment config

CLI Commands

fluxprotocol-cli create <name>   # Scaffold connector
fluxprotocol-cli dev <file>      # Dev with hot reload
fluxprotocol-cli build <file>    # Production build

License

MIT © 2025