DEV Community

Nex Tools
Nex Tools

Posted on • Originally published at nextools.hashnode.dev

How I Built a Full Content Pipeline with Claude Code (No Team, No Budget)

I run a small e-commerce brand. One person. No marketing team. No content budget.

Six months ago, I was posting once a week - maybe. Now I publish daily across 5 platforms with consistent branding, scheduled posts, and automated cross-posting.

The entire system runs on Claude Code + a handful of MCP servers. Here's exactly how.

The Problem: Content Takes Forever

The math was brutal:

  • 1 Instagram carousel: 2-3 hours (research, copy, design, caption, hashtags)
  • 1 blog post: 3-4 hours
  • Cross-posting to 5 platforms: 1 hour of manual formatting

That's 6-8 hours for a single piece of content across all channels. At 5 posts per week, I'd need a full-time content person. I couldn't afford one.

The Architecture: Skills + MCP + Scripts

Claude Code's skill system is the backbone. Here's the stack:

Research Layer:    WebSearch + NotebookLM + Apify MCP
Creation Layer:    Custom skills (content-factory, copywriter, art-director)
Visual Layer:      Gemini MCP (image gen) + Puppeteer (HTML to PNG)
Publishing Layer:  Platform APIs (IG, Hashnode, DEV.to, Blogger, Medium)
Scheduling Layer:  Cron-based scheduled tasks
Enter fullscreen mode Exit fullscreen mode

Each layer has specialized workers - Claude Code skills that know exactly what to do.

Step 1: Research Skills

I have research skills that scan competitors, trending topics, and viral content patterns. They output structured markdown files with hooks, angles, and data points.

The key insight: research and creation are separate skills. A researcher doesn't create. A creator doesn't research. This separation is what makes the quality consistent.

Step 2: Content Factory

A daily skill reads all research files, checks what's already been published (via a publish log), and generates:

  • 10 quote posts (HTML with exact brand styling)
  • 10 reel scripts (JSON with scene-by-scene timing and animations)
  • 10 article drafts (Markdown with frontmatter for each platform)

Everything includes source attribution back to the research that inspired it.

Step 3: Visual Pipeline

Quote posts are HTML files styled with exact brand colors, fonts, and layout. A Puppeteer script converts them to PNG:

node carousel-to-png.js quotes.html output-folder
Enter fullscreen mode Exit fullscreen mode

1080x1080 PNGs, ready for Instagram. No Canva. No Figma. No design skills needed.

Step 4: Platform-Specific Publishing

Each platform has its own publishing script:

  • Instagram: Graph API for carousels and reels
  • Hashnode: GraphQL API
  • DEV.to: REST API with canonical URLs
  • Blogger: Google OAuth2 + Blogger API v3
  • Medium: Chrome-based flow (API closed since 2023)

Step 5: Scheduling

Claude Code's scheduled tasks run the pipeline daily:

09:00 - Content Factory generates daily batch
10:00 - Publisher distributes to all platforms
11:00 - Instagram post goes live
20:00 - Inventory check + next day planning
Enter fullscreen mode Exit fullscreen mode

Results After 30 Days

  • Before: 1 post/week, manual, inconsistent branding
  • After: 5+ posts/day across 5 platforms, consistent, automated
  • Time spent: ~30 minutes/day reviewing and approving
  • Cost: $0 (Claude Code subscription + free API tiers)

What I'd Do Differently

  1. Start with one platform, not five. Should have nailed Instagram first, then expanded.
  2. Quality gates matter. Now I have QA skills that review before publishing.
  3. Research is the bottleneck, not creation. 14 research documents that never became content taught me this.

The Stack

Component Tool Cost
AI Engine Claude Code (Opus) Subscription
Image Gen Gemini MCP Free tier
HTML to PNG Puppeteer + Node.js Free
IG Publishing Meta Graph API Free
Blog Publishing Hashnode/DEV.to/Blogger APIs Free
Scheduling Claude Code scheduled tasks Included
Research WebSearch + Apify $5 credit

Total monthly cost: Claude Code subscription only.


Building something similar? I'm documenting the entire process. Follow for the next post on how the visual pipeline works.

Top comments (0)