Multi-agent research crew: crewai + gpt-researcher, all free from the catalog

I do a lot of domain research for client reports — pulling together papers, blog posts, and documentation into structured summaries. It used to take me 3–4 hours per topic. I've got it down to about 20 minutes of human time using a two-agent crew I assembled from free catalog listings.

The listings

• crewai-multi-agent — the agent orchestration layer; clean role/task/tool abstraction
• gpt-researcher-ai-research — autonomous research agent that searches, reads, and summarises sources

Both free on the catalog. I appreciated that both listings show the full security scan breakdown — when something is going to be making outbound web requests on my behalf, I want to see what the egress analysis found.

The crew setup

from crewai import Agent, Task, Crew
from gpt_researcher import GPTResearcher
import asyncio

# Research agent wraps gpt-researcher
researcher = Agent(
    role="Research Analyst",
    goal="Find and synthesise authoritative sources on a given topic",
    backstory="Expert at finding signal in noisy information landscapes",
    verbose=True
)

# Editor agent cleans and structures
editor = Agent(
    role="Technical Editor",
    goal="Turn raw research into a structured, readable report",
    backstory="Turns jargon-heavy drafts into clean executive summaries",
    verbose=True
)

async def run_research(topic):
    rpt = GPTResearcher(query=topic, report_type="research_report")
    raw = await rpt.conduct_research()
    return await rpt.write_report()

research_task = Task(
    description="Research: {topic}",
    agent=researcher,
    expected_output="A raw research dump with sources"
)

edit_task = Task(
    description="Structure the research into a 500-word executive summary",
    agent=editor,
    expected_output="A clean, structured summary with a bibliography"
)

crew = Crew(agents=[researcher, editor], tasks=[research_task, edit_task])
result = crew.kickoff(inputs={"topic": "retrieval-augmented generation for enterprise search"})

What I actually get

For a typical query like "best practices for MCP server security", the crew:

1. GPT-researcher finds ~12 sources (papers + blog posts + docs)
2. Generates a 1,500-word raw synthesis with inline citations
3. The editor agent restructures it into a 500-word executive summary + bibliography

Total time: ~4 minutes. Total cost: $0 for the tools (I use a local LLM backend via Ollama for the agent reasoning, also free from the catalog).

The quality isn't perfect — it occasionally hallucinates a citation — but for a first-pass research briefing it's genuinely useful. And the entire tool chain came from the agentic category of the catalog at zero cost.