Workflow

Multi-Agent SEO Blog Pipeline that Publishes to Ghost

A CrewAI editorial crew of six specialized agents that researches a topic with live web search, writes a long-form article, optimizes it for SEO, reviews quality, and publishes a draft straight to Ghost CMS.

What This Builds

This recipe builds a code-first content factory: a CrewAI crew of six specialized agents that behaves like a real editorial team. One agent researches the topic against the live web, one writes a ~3,500-word draft, one does keyword research and SEO optimization, one reviews quality, and the pipeline publishes a draft post directly to Ghost CMS. The author reports articles produced in 5-10 minutes at roughly $0.10-0.20 per post with GPT-4o-mini, or $0 when pointed at a local model.

The Stack

  • CrewAI orchestrates the six agents, their roles, tasks, and tool access. Tasks can depend on each other to form the editorial workflow.
  • Brave Search API gives the research agent real-time, current web results instead of stale model knowledge.
  • Ghost CMS receives the finished article as a draft post via its Admin API, formatted and tagged.
  • LLM provider is pluggable: OpenAI | GPT-4o-mini for cheap cloud runs, Claude, or a local Ollama model for zero marginal cost.

Step-by-Step Outline

  1. Define six CrewAI agents, each with a role, goal, and backstory: Research, Content Writer, SEO Optimizer, Quality Reviewer, plus formatting/publishing roles.
  2. Give the Research agent a Brave Search tool so it can pull current sources for the chosen topic.
  3. Define sequential tasks with dependencies: research feeds the writer, the writer’s draft feeds the SEO optimizer, the optimizer’s output feeds the reviewer.
  4. Have the SEO agent generate target keywords, a meta description, and tags from the draft.
  5. Run the crew on a topic; it produces a structured long-form article in minutes.
  6. Publish the reviewed article as a draft to Ghost CMS through the Admin API for a human to do a final pass before going live.

Why This Shape Works

Specialization beats a single “write me a blog post” prompt. Each agent owns one responsibility, which keeps every step inside its context window and makes the pipeline debuggable — if SEO output is weak, you know exactly which agent and prompt to fix. Publishing as a draft rather than live keeps a human in the final loop, which matters for brand and factual accuracy.

Source

Christian Mendieta, “CrewAI Blog Automation: Building a Multi-Agent Content Creation System with Python”: https://christianmendieta.ca/crewai-blog-automation-building-a-multi-agent-content-creation-system-with-python/