By January 2026, if you're building an AI-native business workflow, there's a 70% chance you're using or evaluating CrewAI. With over 44,000 GitHub stars and production deployments across content pipelines, research automation, and customer support systems, it's the fastest path from "I want multi-agent AI" to production.

This tutorial builds a real workflow: a three-agent content research and writing pipeline. No toy examples. Full code. Production patterns included.

What is CrewAI?

CrewAI is a Python framework for multi-agent orchestration. You define Agents (with a role, goal, and backstory), assign them Tools (web search, code execution, database queries, MCP tools), group them into a Crew, and define Tasks that the crew executes sequentially or in parallel.

Unlike LangGraph (which gives you a state machine graph), CrewAI abstracts orchestration behind role-based collaboration. Agents can delegate to each other. The framework handles context passing between tasks.

Prerequisites

Python 3.10+
An OpenAI or Anthropic API key (or any LiteLLM-supported model)
About 15 minutes

Step 1 — Install CrewAI

pip install crewai crewai-tools

Step 2 — Define your agents

Each agent has a role (what it is), a goal (what it's trying to achieve), and a backstory (context that shapes its behaviour). These feed directly into the agent's system prompt.

Create three agents: a Senior Research Analyst with web search tools, a Technical Content Writer, and a Senior Editor.

Step 3 — Define tasks

Tasks describe what needs to be done, who does it, and what output is expected. The expected_output field is important — it tells the agent what format to return.

Create three tasks: research_task (assigned to researcher), writing_task (assigned to writer, with research as context), and editing_task (assigned to editor, with writing as context).

Step 4 — Assemble the crew and run

from crewai import Crew, Process

crew = Crew(

agents=[researcher, writer, editor],

tasks=[research_task, writing_task, editing_task],

process=Process.sequential,

verbose=True,

)

result = crew.kickoff(inputs={"topic": "Model Context Protocol MCP servers"})

Step 5 — Add parallel execution for speed

For independent tasks, use Process.hierarchical with a manager agent. The manager delegates work and ensures quality without you having to micromanage the orchestration.

Production patterns

Wrap in FastAPI for async execution. Connect to MCP servers as tools using MCPTool from crewai_tools. Add memory with the Memory class so agents retain context across sessions.

FAQ

What is CrewAI used for? Multi-agent AI workflows — content generation, research automation, lead enrichment, customer support systems.

CrewAI vs LangGraph — which should I use? CrewAI for role-based coordination and fast shipping. LangGraph for fine-grained state control, complex loops, and human-in-the-loop checkpoints.

Does CrewAI work with any LLM? Yes — any LiteLLM-compatible model: GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, Llama 3, Mistral.

Can CrewAI agents use MCP servers? Yes, via MCPTool in the crewai_tools package.

How do I deploy a CrewAI workflow to production? As a FastAPI endpoint, Celery background worker, AWS Lambda container, or via CrewAI Cloud.

CrewAI Tutorial 2026: Build a Multi-Agent Workflow in 30 Minutes