h/t the original builder, a startup founder (ex-DevRel) who open-sourced this workflow and shared it in a Reddit post.
TL;DR
- The founder was paying around $500 for content and still did much of the coordination work.
- The open-source workflow runs from RSS monitoring to drafting, SEO checks, human review, and a GitHub PR.
- It learns from your existing blog to build writing rules instead of writing generic articles.
- The author reports 30+ articles and around 10,000 impressions in two weeks; clicks were not shared.
The problem it solves
The founder says their startup was paying around $500 for content and ghostwriting support. Even so, a good chunk of the work kept coming back to them:
- Coordinating with writers
- Explaining the industry and product over and over
- Finding relevant topics and researching SEO keywords
- Checking sources
- Maintaining a consistent writing style, which the founder calls the toughest part
There's also a timing issue. Writers aren't monitoring RSS feeds, Reddit, or industry discussions all day. When a competitor launches something or a hot discussion starts, by the time you brief someone and get a draft, the conversation may have moved on.
How the workflow works
The founder built an MCP-based content pipeline for their startup, Wayzyy, and later open-sourced it. The full flow:
RSS / News → Research → Keywords → Learn from existing articles → Rulebook → Outline → Draft → SEO Check → Human Review → Reddit Ideas → GitHub PR
Key features
- Learns from your corpus: instead of generating every article from scratch, it studies previously published content and builds writing rules from it.
- Keyword management: it manages keyword ideas using TinyFish.
- SEO checks: it checks headings, links, FAQs, and metadata, and generates FAQ JSON-LD separately for publishing.
- One shared Python service: a dashboard, Node.js scripts, and MCP-compatible assistants like Claude or Cursor all call the same underlying methods.
- Visual builder: a drag-and-drop interface connects research, RSS monitoring, keyword planning, writing rules, drafting, SEO checks, human approvals, Reddit ideas, and GitHub publishing.
- Client workspaces: separate workspaces let agencies keep each client's style and requirements apart.
The architecture choice
To avoid writing the logic several times, the founder used Graftcode's gateway to load the Python service, generate the client used by the dashboard and scripts, and expose those methods through a built-in MCP endpoint. The founder says that meant no separate API routes, request models, client wrappers, or second MCP server for each interface. An earlier hand-built MCP bridge is kept in the repo's legacy folder for comparison.
Disclosure worth noting: the founder says the project participates in the Graftcode Impact Programme and receives rewards depending on the traction it generates.
Results so far
According to the founder, the team has used the workflow across 30+ articles and saw around 10,000 impressions in the last two weeks.
Treat that number carefully. One commenter in the thread asked how many of those impressions turned into clicks or signups, noting that impressions across 30 new articles can look big while click-through stays tiny. The founder didn't share click or conversion data, so the real business impact is still unproven.
Manual process vs. this workflow
| Task | Hiring writers | MCP content workflow |
|---|---|---|
| Topic discovery | Depends on you or the writer | RSS and news monitoring step |
| Keyword research | Manual | Keyword ideas managed in the pipeline |
| Brand voice | Re-explained to each writer | Rulebook built from your existing articles |
| SEO hygiene | Manual review | Automated checks on headings, links, FAQs, metadata |
| Quality control | Editor or founder | Human review step built in |
| Cost | The founder paid around $500 | Free and open source, plus your own setup time and AI usage |
Limitations and caveats
- Technical setup. This is a repo with a Python service, Node.js scripts, and MCP integrations. It's not a polished SaaS, so expect to spend time getting it running.
- Needs good examples. The founder says drafts get closer to your style once the system has enough good examples to learn from. A brand-new blog won't benefit as much.
- Be careful with Reddit. A commenter warned that subreddits are quick to spot posts that came out of a content pipeline, which can burn an account fast. They suggested keeping that step fully human, which is sound advice.
- Pipeline order. Another commenter noted that the Reddit Ideas step only runs after an article is reviewed, which works against the timing goal. They suggested branching it straight off the RSS step so live threads get a same-day response.
- Impressions aren't revenue. Track clicks, signups, and sales before you promise clients anything.
How to try it
- Find the open-source repo linked from the founder's post and read the README and walkthrough video.
- Set up the Python service locally and connect it to Claude or Cursor through the MCP endpoint.
- Feed it at least a handful of your best existing articles so it can build a rulebook.
- Add 3-5 RSS feeds from your niche and let it surface topic ideas.
- Generate one outline and draft, then edit it heavily and compare the output to your normal process.
- Keep the human review step mandatory, and do any Reddit posting yourself.
- Track clicks and signups in Google Search Console and analytics, not just impressions.
- Once you trust the output, create a separate workspace for one pilot client.
Originally published on SoloAgentFlow, which also covers three ways freelancers and agencies could build a service around it.
Top comments (0)