YouTube to Blog Post with Claude Code

Key Takeaways

• One command, one blog post. A Claude Code skill can take a YouTube URL, extract the transcript, and produce a fully optimized article with internal links, meta tags, and embedded video
• MCP publishes directly to WordPress, Ghost, or any headless CMS without copy-pasting between tools
• 49% of marketers say they don't repurpose enough despite knowing they should (Optimizely, 2024)
• Transcript is the base, not the output. The skill restructures content for reading, not transcription. Header hierarchy, keyword density, and link placement happen automatically
• WordPress MCP is official. WordPress.org shipped the MCP Adapter in February 2026, making AI-to-CMS publishing a first-party feature
• Start with the SEO Command Center skill setup if you haven't configured Claude Code for SEO work yet

Every SEO team has the same problem: video content performs well on YouTube but never becomes a blog post. The transcript sits in the creator's head (or a Descript export), and turning 15 minutes of spoken content into a structured article requires reformatting, keyword optimization, and CMS wrangling that most teams skip.

Content repurposing ranks as the fifth most popular digital marketing trend of 2026 (HubSpot State of Marketing, 2026). Yet 49% of marketers say they don't do enough of it (Optimizely, 2024). The bottleneck isn't strategy. It's execution.

The infrastructure to fix this now exists. MCP servers grew from roughly 100 in November 2024 to over 5,800 by mid-2025, with 8 million total downloads (mcpevals.io, June 2025). YouTube transcript MCP servers, CMS publishing adapters, and Claude Code skills can chain together into a single command that turns a video URL into a draft blog post.

This guide shows you how to build a Claude Code skill that takes a YouTube URL and produces a published blog post draft. We'll cover transcript extraction, the SEO skill structure, MCP publishing setup, and the gotchas that'll waste your time if you don't know about them upfront.

How the Pipeline Works

The workflow has three stages, each handled by a different tool in Claude Code's environment:

YouTube URL
    ↓
[Transcript Extraction]  ← MCP tool or API
    ↓
[SEO Blog Writer Skill]  ← Your SKILL.md file
    ↓
[CMS Publishing]          ← WordPress/Ghost MCP
    ↓
Published Draft

You paste a YouTube URL into Claude Code along with a target keyword. Claude Code extracts the transcript, applies your SEO skill's formatting rules, and pushes the result to your CMS as a draft. One prompt, three tools, zero tab-switching.

What You Need Before Starting

Requirement	Detail
Claude Code	Claude Pro ($20/mo) works for testing. Claude Max ($100/mo) for production volume
Transcript tool	YouTube Transcript MCP server (490+ stars option), Python API, or Supadata
SEO skill file	A SKILL.md that defines your article structure, tone, keyword rules
CMS with MCP	WordPress (official MCP Adapter), Ghost, or custom API integration

Step 1: Set Up Transcript Extraction

Claude Code can't watch a YouTube video. It needs the transcript as text. There are several ways to get it.

Option A: YouTube Transcript API (Free, Self-Hosted)

The youtube-transcript-api Python package pulls transcripts directly from YouTube's auto-generated or manual captions. No API key needed.

pip install youtube-transcript-api

Create a simple fetcher script:

from youtube_transcript_api import YouTubeTranscriptApi

def get_transcript(video_id: str) -> str:
    transcript = YouTubeTranscriptApi.get_transcript(video_id)
    return " ".join([entry["text"] for entry in transcript])

Tell Claude Code to run this script when it encounters a YouTube URL. Add it to your project's CLAUDE.md:

## YouTube Transcripts
When given a YouTube URL, extract the video ID and run:
python3 scripts/get_transcript.py <video_id>
Use the output as the source material for blog content.

Option B: MCP-Based Transcript Servers

Over 40 YouTube transcript MCP servers exist as of early 2026. The best options for Claude Code:

MCP Server	Stars	Best For
kimtaeyoon83/mcp-server-youtube-transcript	490+	Multi-language support, ad filtering, Shorts
hancengiz/youtube-transcript-mcp	--	Built for Claude Code, has dedicated agent guide
jkawamoto/mcp-youtube-transcript	--	Long videos (cursor pagination for 50K+ chars)
ergut/youtube-transcript-mcp	--	Remote server, zero local install
Supadata MCP	--	Multi-platform (YouTube, TikTok, Instagram)

Add one to your Claude Code MCP config:

{
  "mcpServers": {
    "youtube-transcript": {
      "command": "npx",
      "args": ["-y", "youtube-transcript-mcp"]
    }
  }
}

The MCP approach is cleaner because Claude Code calls the tool natively without shelling out to a Python script. One constraint: MCP responses have a 25,000 token limit. For videos over 30 minutes, use a server with pagination support (jkawamoto).

Option C: Manual Transcript Paste

For low volume, grab the transcript from YouTube's UI (click "Show transcript" under any video) and paste it into your prompt. Less elegant, works fine for 5 articles a month.

Step 2: Build the SEO Blog Writer Skill

This is where the quality happens. A Claude Code skill (SKILL.md file) defines every structural rule your blog post needs to follow. Without it, you get a transcript reformatted into paragraphs. With it, you get a publishable article.

The Skill File Structure

Create skills/content/seo-blog-writer/SKILL.md:

# SEO Blog Post Writer

## Trigger
/blog-from-video [YouTube URL] --keyword "[target keyword]"

## Instructions

### Source Processing
1. Extract transcript from the provided YouTube URL
2. Identify the core topic, key arguments, and supporting points
3. DO NOT transcribe. RESTRUCTURE for reading.
   Spoken content is messy: tangents, filler, repeated points.
   Reorganize into a logical article flow.

### Article Structure
1. H1: Include target keyword. Under 60 characters.
2. Opening paragraph: Lead with a stat or specific claim.
   Target keyword in first sentence. 2-3 paragraphs max.
3. H2 sections: Each covers one major topic from the video.
   Use secondary keywords in H2s naturally.
4. H3 subsections: For step-by-step or list content.
5. Code blocks: Include any commands or code shown in video.
6. FAQ section: 4-6 questions targeting long-tail queries.

### SEO Rules
- Target keyword in: H1, first paragraph, one H2, meta
  description, URL slug
- Keyword density: 1-2% for primary, 0.5-1% for secondary
- Internal links: 2-3 to existing site content (check sitemap)
- External links: 2-4 to authoritative sources, open in new tab
- Word count: 1,200-2,000 words (not padding, real content)
- Meta description: Under 155 characters, includes keyword

### Media
- Embed the source YouTube video in the article body
- Add alt text to any images referencing the section context

### Tone and Style
- Technical peer, not beginner tutorial voice
- Short paragraphs (2-4 sentences)
- Bold key terms on first mention per section
- No filler intros, no "In this article we'll cover..."

Why the Skill File Matters

The skill file is training data for your specific output. Every site has different heading conventions, CTA placements, internal linking patterns, and tone. Without these rules codified, Claude Code produces generic AI blog content that reads like every other AI-generated post.

Two critical rules to get right:

1. "Do not transcribe. Restructure for reading." This single instruction is the difference between a readable article and a cleaned-up transcript. Spoken content has a different information architecture than written content. Your skill needs to force the reorganization.

2. Internal linking. Tell the skill to check your sitemap or a list of existing posts and link to them contextually. This turns isolated blog posts into a linked content hub.

Step 3: Connect Your CMS via MCP

MCP (Model Context Protocol) lets Claude Code interact with external services as native tools. For publishing, this means Claude Code can create a post, set the category, add meta descriptions, and upload images without you touching the CMS dashboard.

WordPress (Official MCP Adapter)

WordPress shipped official MCP support in February 2026 (WordPress Developer Blog). Two paths:

WordPress.com (hosted):

{
  "mcpServers": {
    "wordpress": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/wordpress-mcp"]
    }
  }
}

WordPress.com uses OAuth 2.1 for authentication. The setup flow opens your browser, you authorize, and you're connected. (WordPress.com MCP docs)

WordPress.org (self-hosted):

Install the MCP Adapter plugin, then configure the local proxy in your Claude Code MCP settings. The adapter exposes WordPress Abilities (create post, set categories, upload media) as MCP tools.

What Claude Code Can Do via WordPress MCP

Once connected, Claude Code gets access to these operations:

MCP Tool	What It Does
create_post	Creates a new post with title, content, status (draft/publish)
set_categories	Assigns post to existing categories
set_tags	Adds tags to the post
upload_media	Uploads images and sets featured image
update_meta	Sets meta description, focus keyword (Yoast/RankMath)
set_permalink	Customizes the URL slug with target keyword

Other CMS Options

Every major CMS now has MCP support. The landscape as of March 2026:

CMS	MCP Server	Maintainer
Ghost	MFYDev/ghost-mcp	Community
Sanity	sanity-io/sanity-mcp-server	Official (remote hosted)
Contentful	contentful/contentful-mcp-server	Official (Beta)
Webflow	webflow/mcp-server	Official (first CMS to ship MCP, April 2025)
Payload CMS	Official MCP Plugin	Official
Strapi	strapi-mcp	Community

For Next.js + MDX sites (like this one), a Blog Publisher MCP Server can commit .mdx files with frontmatter directly to your GitHub repo, triggering a Vercel rebuild. No CMS needed.

The "Draft First" Rule

Always publish as draft, not live. This is non-negotiable for production use. The skill produces good output, but every post needs a human pass before going live. Check:

• Factual accuracy (LLMs can misrepresent video content)
• Internal links actually point to real pages
• The meta description reads well to a human, not keyword-stuffed
• Code blocks are syntactically correct

Step 4: Run the Full Pipeline

With transcript extraction, the SEO skill, and CMS MCP configured, the entire workflow is one prompt:

/blog-from-video https://youtube.com/watch?v=VIDEO_ID
--keyword "claude code seo workflow"
--secondary "seo automation, claude code skills"
--title "How to Automate SEO Workflows with Claude Code"

Claude Code will:

1. Extract the transcript (via MCP tool or Python script)
2. Analyze the content structure and identify key topics
3. Apply your SEO skill rules to restructure into an article
4. Generate internal links by checking your sitemap
5. Create the post in WordPress as a draft
6. Set category, tags, meta description, permalink, and featured image

Total time from prompt to draft: 30-90 seconds depending on video length and model speed.

Optimizing for Speed

• Use Sonnet 4.6 instead of Opus for this workflow. The quality difference is negligible for structured content generation, and you'll use fewer tokens
• Turn off extended thinking. The skill file provides enough structure that the model doesn't need extra reasoning time
• Set up auto-approve for the MCP tools you trust (file writes, WordPress draft creation). This eliminates the permission prompts that slow down the workflow

What This Gets You (and What It Doesn't)

What works well

Volume. 10 YouTube videos become 10 blog posts in under 15 minutes of active work. For agencies managing client YouTube channels, this is a content multiplication play.

SEO coverage. A video ranking on YouTube and a blog post ranking on Google cover two search surfaces from one piece of content. 22% of Google searches return at least one YouTube video on page one (SearchAtlas, 2025). Embedding the source video increases on-page dwell time, which reinforces the page's quality signal. 93% of marketers report positive ROI from video, the highest in 12 years of Wyzowl tracking (Wyzowl, 2025).

Consistency. The skill file enforces your SEO standards on every post. No more "the intern forgot the meta description" or "this post has no internal links." The rules execute every time.

What doesn't work

Original analysis. The blog post can only contain what the video covered. If the video is surface-level, the blog post will be surface-level. The skill can restructure content, not create new insights.

Perfect accuracy. Claude Code occasionally misinterprets what was said in the video, especially with technical jargon or numbers. The human review pass catches these.

High-competition keywords. A repurposed video transcript won't outrank a purpose-built 3,000-word guide with original research, custom graphics, and 50 backlinks. This workflow is for capturing long-tail and mid-competition terms where good-enough content wins.

Scaling This for Agency Use

For agencies managing multiple client YouTube channels, the workflow compounds:

1. One skill file per client with their brand voice, internal linking map, and CMS credentials
2. Batch process weekly: pull all new client videos, run the skill against each, review drafts in bulk
3. Track results in GSC: compare click/impression growth on pages created via this pipeline vs. manually written content
4. Iterate the skill based on which posts perform. If certain heading structures or content patterns correlate with higher rankings, encode those patterns into the skill file

The skill file becomes a living document that gets better with every feedback loop. After 20-30 posts, your SEO blog writer skill produces output that needs minimal editing.

FAQ

Can I use this with any YouTube video, or only my own?

You can extract transcripts from any public YouTube video with captions enabled. The ethical question is different from the technical one. Repurposing your own video content is straightforward. Repurposing someone else's video into your blog post is content you didn't create. If you're using competitor or industry videos as source material, the output needs substantial original analysis layered on top, not a reformatted transcript.

What if the YouTube video doesn't have captions?

About 73% of YouTube videos have auto-generated captions. If a video doesn't have any, you'll need a separate transcription service (Whisper API, AssemblyAI, or Descript) to generate the transcript first, then feed it to the Claude Code skill.

How is this different from ChatGPT's "summarize this video" feature?

ChatGPT can summarize a video, but the output isn't structured for SEO. There's no keyword targeting, no header hierarchy optimization, no internal linking, no CMS publishing. Claude Code skills give you a repeatable, rule-based pipeline that produces consistent SEO-formatted output and pushes it directly to your site.

Does embedding the YouTube video in the blog post help SEO?

Embedding video increases average time on page. Pages with embedded video see 2-3x longer dwell time than text-only equivalents. Google doesn't directly rank pages higher for having video, but the engagement signals (time on page, lower bounce rate) contribute to overall page quality assessment.

What Claude plan do I need for this?

Claude Pro ($20/month) works for testing and low volume (5-10 posts/month). For production use at agency scale, Claude Max ($100/month) gives you the headroom. The transcript extraction + skill execution + MCP publishing chain uses more tokens than a simple chat, so budget accordingly.

Can I customize the output format for different clients?

Yes. Create separate skill files per client: skills/content/client-a-blog-writer/SKILL.md and skills/content/client-b-blog-writer/SKILL.md. Each skill encodes the client's voice, keyword strategy, CMS connection, and formatting rules. Switch between them by invoking the specific skill name.