​
Providers Overview

Kubiya Workflow SDK supports multiple types of providers to enable flexible workflow orchestration and integration with various AI systems. This section covers two main categories of providers:

​
Provider Categories

🤖 Agent Servers

Full-featured orchestration engines that provide complete workflow management, execution, and AI-powered automation

🔌 MCP Providers

Model Context Protocol providers that enable integration with AI assistants like Claude Desktop, Cursor, and custom agents

​
Agent Servers

Agent servers are complete orchestration engines that handle the full lifecycle of workflow execution. They provide:

​
Core Capabilities

  • Workflow Generation: AI-powered creation of workflows from natural language
  • Intelligent Execution: Context-aware step-by-step execution with error handling
  • Multi-Agent Orchestration: Coordination of specialized agents for complex tasks
  • Real-time Streaming: Live feedback and progress updates during execution

​
Available Agent Servers

Google Agent Development Kit (ADK)

Google’s comprehensive agent framework providing intelligent workflow orchestration, multi-agent coordination, and seamless integration with Google Cloud services.

​
Architecture Overview

​
MCP Providers

MCP (Model Context Protocol) providers enable integration with AI assistants and development tools. They bridge the gap between external AI systems and Kubiya’s workflow capabilities.

​
What is MCP?

The Model Context Protocol is an open protocol that standardizes how AI assistants connect to external tools and data sources. It enables:

  • Tool Discovery: AI agents discover available workflow capabilities
  • Context Sharing: Share relevant workflow context with AI assistants
  • Action Execution: AI can trigger and monitor workflow executions
  • Real-time Integration: Stream workflow results back to AI interfaces

​
Available MCP Providers

FastMCP

Native Python implementation of the Model Context Protocol, providing seamless integration with Claude Desktop, Cursor, and other MCP-compatible tools.

​
MCP Integration Flow

​
Choosing the Right Provider

Use CaseRecommended ProviderWhy
Full AI Automation PlatformAgent Servers (ADK)Complete workflow lifecycle management
AI Assistant IntegrationMCP Providers (FastMCP)Seamless integration with existing AI tools
Enterprise OrchestrationAgent Servers (ADK)Advanced multi-agent coordination
Developer Tool IntegrationMCP Providers (FastMCP)IDE and development workflow integration
Google Cloud IntegrationAgent Servers (ADK)Native Google services integration
Cross-Platform AI ToolsMCP Providers (FastMCP)Protocol standardization

​
Execution Models

​
Serverless Containerized Execution

Both provider types leverage Kubiya’s containerized execution model:

🐳 Universal Container Runtime

Every workflow step runs in its own Docker container, providing:

  • Complete Software Freedom: Install and run any software, library, or tool
  • Language Agnostic: Python, Node.js, Go, Rust, Java - use any language
  • Stateless Execution: Each execution starts fresh, ensuring consistency
  • Infinite Scalability: Workflows scale automatically based on demand
  • Resource Isolation: Each step runs in its own secure environment

​
Example: Multi-Language Pipeline

from kubiya_workflow_sdk import Workflow, Step

# Create a workflow that uses multiple technologies
workflow = Workflow(
    name="multi-tech-pipeline",
    runner="kubiya-hosted"
)

# Data extraction with Python
workflow.add_step(Step(
    name="extract-data",
    image="python:3.11",
    packages=["requests", "pandas"],
    code="""
    import requests
    import pandas as pd
    
    # Extract data from API
    response = requests.get('https://api.example.com/data')
    df = pd.DataFrame(response.json())
    df.to_csv('extracted_data.csv')
    """
))

# Analysis with R
workflow.add_step(Step(
    name="statistical-analysis",
    image="r-base:latest",
    code="""
    library(tidyverse)
    
    # Perform statistical analysis
    data <- read.csv('extracted_data.csv')
    analysis <- summary(lm(y ~ x, data))
    write.csv(analysis$coefficients, 'analysis_results.csv')
    """
))

# Visualization with Node.js
workflow.add_step(Step(
    name="create-dashboard",
    image="node:20",
    packages=["d3", "express"],
    code="""
    const express = require('express');
    const d3 = require('d3');
    
    // Create interactive dashboard
    const app = express();
    // Build visualization from analysis results
    """
))

​
Next Steps

🤖 Agent Servers Guide

Learn about full-featured orchestration with Google ADK and other agent servers

🔌 MCP Integration

Integrate workflows with AI assistants using FastMCP

🚀 Frontend Integration

Build web applications that consume workflow providers

📚 End-to-End Tutorial

Complete tutorial: Local development to production deployment