How I Built a Claude Code Content Pipeline That Publishes Daily

I publish a new blog post almost every day now. Not because I spend all day writing — I don't — but because I built a Claude Code content pipeline that handles the mechanical parts of content production while I focus on strategy and editing. The system researches keywords, generates first drafts, optimizes meta tags, and prepares files ready to deploy. What used to take 3–4 hours per post now takes about 45 minutes of actual human time.

This isn't theoretical. The post you're reading right now came out of this pipeline. So did the last 30 posts on this blog. And the results are real: organic traffic to alejandroarce.com has doubled in four months, and I'm ranking on page one for "Claude Code consultant Vancouver" and a dozen related terms.

Here's exactly how the Claude Code content pipeline works, what parts are automated, and where I still make judgment calls by hand.

The Five Stages of the Pipeline

My content pipeline has five distinct stages. Each one is either fully automated or requires a quick human review before moving to the next step. The stages are:

Keyword research and topic selection — identifying what to write about based on search volume and competitive gaps
Outline generation — turning a keyword into a structured content brief with H2s, suggested word count, and internal linking opportunities
Draft creation — writing the full post based on the outline, optimized for the target keyword
SEO optimization — generating meta tags, checking keyword density, ensuring all on-page signals are correct
File preparation and deployment — creating the final HTML file with proper schema markup, nav structure, and internal links

The entire flow runs inside Claude Code sessions, with each stage handing off structured data to the next. The only points where I intervene are approving the topic list (stage 1), reviewing the outline (stage 2), and doing a final edit on the draft (stage 3). Everything else is automated.

Stage 1: Keyword Research with MCP

The pipeline starts with keyword research. I use Claude's Model Context Protocol (MCP) to connect to SEMrush and Ahrefs APIs. A Claude Code script pulls keyword data for my target niche — in my case, "Claude Code," "AI consulting," "Vancouver marketing automation," and related terms.

The script filters for keywords with:

Search volume between 50 and 2,000 per month (the sweet spot for a local consultant)
Keyword difficulty under 40 (winnable without a massive backlink campaign)
Commercial or informational intent (I skip purely navigational queries)
No existing coverage on my site (to avoid cannibalization)

The output is a ranked list of 20–30 keywords. I review it, remove anything that doesn't align with my actual service offerings, and approve the final batch. That approval step takes about 10 minutes once a week.

Why MCP Matters Here

Before MCP, I had to export CSV files from SEMrush, clean them in Excel, and manually cross-reference them with what I'd already published. It was a 90-minute process. Now the MCP integration handles all of that in a single Claude Code session. The time savings on this step alone justified building the whole pipeline.

Stage 2: Outline Generation

Once I've approved the keyword list, the next script takes each keyword and generates a content brief. The brief includes:

A primary H1 and 2–3 alternative title options
Suggested H2 structure based on what's ranking in the top 5 Google results
Target word count (usually 900–1,300 words for my audience)
Related keywords to include naturally in the body
2–3 internal links to existing posts on my site

The prompt that drives this looks like:

You are a content strategist. For the keyword "[KEYWORD]", analyze the top 5 Google results and create a content brief.

Include:
1. Primary H1 (under 70 chars, keyword in first half)
2. Five H2 headings that cover what users are searching for
3. Target word count
4. 3–5 related keywords to weave in naturally
5. Internal link opportunities from this list: [EXISTING_SLUGS]

Return JSON with keys: h1, h2s, wordCount, relatedKeywords, internalLinks.

I spot-check the outlines to make sure they're not too generic. If an outline looks like it could apply to any AI consultant, I reject it and refine the prompt. The goal is specificity — every post should feel like it's written by someone who actually does this work in Vancouver, because it is.

Stage 3: Draft Creation

This is where the heavy lifting happens. Claude Code takes the approved outline and writes a full draft. The prompt includes:

The target keyword and related keywords
The H1 and H2 structure from the outline
My writing guidelines (first person, Vancouver context, practical tone, no fluff)
The internal links to include
A requirement to include 2–4 actionable takeaways

The output is a 900–1,300 word draft that reads like I wrote it. About 80% of the time, the draft is good enough to publish with only light editing. The other 20% needs more significant rewrites, usually because the examples are too generic or the tone drifts into marketing-speak.

The key to making this work: I feed Claude Code examples of my best-performing posts as reference material. It learns my voice, my structure, and the level of detail I typically include. Without that training data, the drafts would be serviceable but bland.

Stage 4: SEO Optimization

Once the draft is approved, a separate script handles the SEO layer. It generates:

Title tag (under 60 characters, keyword in the first half)
Meta description (130–155 characters, includes a soft CTA)
Open Graph and Twitter card tags
JSON-LD structured data for BlogPosting schema
Canonical URL

It also runs a quick check on keyword density (I aim for 1–2% for the primary keyword) and flags any issues with heading hierarchy or missing internal links. This is the same audit workflow I covered in my post on SEO automation, just applied to new content instead of existing pages.

Stage 5: File Preparation and Deployment

The final stage is file generation. Claude Code takes the optimized draft, wraps it in the site's HTML template, and outputs a complete index.html file ready to drop into the /blog/[slug]/ directory.

The template includes:

Site-wide navigation with mobile menu
Google Analytics tag (G-MT7GNCCZ68)
Breadcrumb navigation
Post hero with category badge, H1, and meta info
Post body with proper heading hierarchy
CTA band linking to my booking calendar
Post navigation footer
Site footer with all standard links

I also generate a "blog card" snippet that gets added to the main /blog/index.html grid. The card includes the post title, excerpt, category, read time, and a link to the full post.

Deployment is still manual — I review the HTML in a browser, make sure internal links work, and then push to GitHub. That takes about 5 minutes. I could automate it with a git commit script, but I like the final check.

What This Approach Can't Do

I want to be clear about the limits. This pipeline works because I've already done the hard strategic work: identifying my niche, understanding what my audience needs, and building a backlog of proven content formats. Claude Code is executing on a strategy I defined.

It's not going to write thought leadership pieces that require original research or interviews. It's not going to produce viral content or deeply personal essays. And it won't replace the judgment calls around which keywords are actually worth pursuing in a competitive market like Vancouver.

What it does incredibly well is produce consistent, SEO-optimized, on-brand content at a volume that would be impossible for me to sustain manually. That's the use case: turning strategy into published assets faster than you could with a human writer.

Results So Far

Since I started using this Claude Code content pipeline in January, I've published 47 posts. Organic traffic is up 112%. I'm ranking on page one for 18 target keywords. Three of those posts have generated inbound leads that turned into paid consulting projects.

The time investment to set up the pipeline was about 12 hours spread over two weeks. The ongoing time cost is roughly 45 minutes per post for review and editing, plus 10 minutes a week to approve new keyword batches. Compare that to the 3–4 hours it used to take me to write and publish a single post manually.

If you're a consultant, agency, or B2B service provider who needs to publish regularly but doesn't have time to write from scratch every week, this approach is worth exploring. And if you want help building something similar for your business, I do this as a service — usually with a working prototype in 48 hours.

Getting Started: Four Practical Steps

If you want to build your own Claude Code content pipeline, here's where I'd start:

Define your content strategy first. What keywords matter? What does your audience need? What formats have worked in the past? Claude Code will execute on this — it won't create it for you.
Start with one stage. Don't try to automate the whole pipeline at once. Begin with draft generation for a single keyword, refine it until quality is consistent, then add the other stages.
Build a library of reference content. Feed Claude Code 5–10 of your best posts so it can learn your voice and structure. The better the training data, the better the output.
Review everything before publishing. Automation should speed up your workflow, not replace your judgment. Every post still needs a human edit.

The tools are here. The question is whether you're willing to invest the upfront time to set them up. For me, it's been one of the highest-leverage projects I've done this year.

If you have questions about how any of this works or whether it makes sense for your situation, the FAQ page covers most of the common ones. And if you want to see other ways I'm using Claude Code for client work, check out what I've built so far.