📝 Blog Writing Agent - Full Stack Project

An AI-powered full-stack application that researches, plans, and drafts blog content using an agent workflow plus external tools.

📖 Table of Contents

• Demo Videos
• Problem This App Solves
• Problems in the Project
• What This Project Does
• Tech Stack
• Architecture Flow
• Project Areas
• Getting Started
• Environment Variables
• Current Status
• Next Milestones
• Planning Source
• Contributing

🎬 Demo Videos

Fast-Paced Highlight Reel

A quicker highlight reel of the main workflow and output.

Demo of Full Project on YouTube

💡 Tip: Click the settings icon (⚙️) in the video player and select Speed → 0.25x to watch at slower speed for better understanding.

Full Long Video

A complete end-to-end walkthrough of the project, including setup, workflow execution, and final output review.

Full Long Demo on YouTube

🚩 Problem This App Solves

Writing a high-quality blog is usually slow and manual. Most people have to:

• Research multiple sources by hand
• Gather and organize facts manually
• Draft, edit, and rework the structure several times

If you only use a simple LLM, it often produces generic content from model memory and cannot reliably do live research, source gathering, and multi-step editorial refinement.

This app solves that by using an agent workflow that can:

• Do live research with tools
• Think in steps (plan before writing)
• Write section by section
• Edit and assemble a full draft
• Run critique and revision loops
• Support human review before generating the final blog

⚠️ Problems in the Project

There are still a few important problems we need to improve:

• The current Ollama-based workflow takes nearly 17 minutes to complete end to end
• Total token usage is very high, at roughly 86,000 tokens including input and output
• We should reduce runtime and token cost to make the system faster and more efficient
• Right now the agent is not a chat app after generation
• After a blog is generated, the user should be able to edit it, add headings, or make other changes using the LLM
• That post-generation editing flow would make the experience feel more like a chat-based assistant

🎯 What This Project Does

The Blog Writing Agent is a full-stack AI system where:

• The frontend collects prompts and presents progress/results.
• The backend runs a LangGraph workflow to coordinate writing steps.
• The MCP server provides tool-based research (search and page fetch).

🏗️ Architecture Flow

User Prompt
	↓
Frontend (Next.js)
	↓
Backend Orchestration (LangGraph/FastAPI layer)
	↓
MCP Tools (web search + fetch)
	↓
Plan → Draft → Edit → Critique → Final Output

🤖 Workflow Nodes (Current Graph Design)

🛠️ Tech Stack

Backend

• Python 3.11+
• LangGraph + LangChain
• FastAPI (route layer present, partially scaffold-level)
• PostgreSQL-oriented checkpointing setup in current agent runner

Tool Server

• FastMCP
• Tavily-based search and page-fetch tooling

Frontend

• Next.js
• React
• TypeScript
• Tailwind CSS

📁 Project Structure

blog-writing-agent/
├── frontend/        # Next.js frontend app
├── backend/         # Agent workflow, nodes, API scaffolding
├── mcp_server/      # FastMCP research tool server
├── docs/            # Exported Notion project docs/tasks
└── README.md

📁 Project Areas

• frontend/: user interface and client-side state
• backend/: workflow orchestration and backend modules
• mcp_server/: FastMCP tools used by the agent
• docs/: exported Notion planning/task documentation

🚀 Getting Started

Prerequisites

• Python 3.11+
• Node.js 18+
• pnpm
• uv (recommended)

1) Run MCP Server

cd mcp_server
uv sync
uv run fastmcp run server.py

Optional SSE mode:

python server.py --sse

2) Run Backend Agent Workflow

cd backend
uv sync
python -m agent.main

3) Run Frontend

cd frontend
pnpm install
pnpm dev

Open: http://localhost:3000

🔐 Environment Variables

Create service-local .env files where needed.

Common variables used across the repo/docs:

• TAVILY_API_KEY
• MCP_AUTH_TOKEN
• DB_URL (or DATABASE_URL in deployment-oriented docs)
• HORIZON_TOKEN
• HUGGINGFACEHUB_API_TOKEN
• NEXT_PUBLIC_BACKEND_URL

📌 Current Status

Implemented now:

• LangGraph workflow with multiple nodes and conditional edges
• MCP tool integration in the agent loop
• Notion-derived planning docs and task structure

Completed recently:

• Backend implementation is completed
• Frontend implementation is completed
• Total time spent on this complete project (browsing + coding), measured with WakaTime: 99 hours 47 minutes

Deployment status (current):

• MCP server is deployed
• Frontend is not deployed yet
• Backend is not deployed yet

Planned upgrades:

• Deploy frontend and backend in future milestones
• Use OpenAI models via API (cloud) in place of the current local Ollama/Qwen 3.5 setup

To avoid confusion, treat this repository as active development rather than a fully completed production release.

🛠️ Next Milestones

• Add richer end-to-end API + UI integration
• Expand structured test coverage across services
• Improve deployment automation and production configuration
• Add publish/export flows for generated blogs
• Add configurable writing styles and tone controls

🗺️ Planning Source

Primary planning source:

• Notion project: https://southern-dogsled-629.notion.site/Blog-Writing-Agent-Full-Stack-Project-32fc1cfa959c8192abaffdc427593c41
• Exported docs in this repo: docs/

🤝 Contributing

1. Create a feature branch.
2. Keep changes focused by service.
3. Update docs for behavior changes.
4. Open a PR with testing notes.