How to use Codex to build a publishing MCP server
Ray de Guzman used Codex and Claude planning to build a Shopify publishing MCP server, reviewing phase reports while Codex built, tested and deployed tools.
nomaditsu.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build an MCP publishing interface for [content platform] using [content schema and taxonomy]. Work in reviewed phases for authentication, tools and deployment. After each phase, report implemented behavior, tests and remaining risks. Require confirmation for [destructive actions] and wait for phase acceptance.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Ray de Guzman
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- 16-tool publishing server built; article itself published/edited through pipeline.
- Human involvement
- Author confirmed each phase and directed architecture; hard deletes are confirm-gated.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Claude in Notion planned self-contained phases.
- Author passed prompts to Codex.
- Codex wrote tests/code, fixed failures, deployed and produced progress reports.
Ray de Guzman built a publishing MCP server so content drafted in his usual tools could reach his Shopify website without a manual admin workflow. He used Claude in Notion to plan the architecture, API surface and rollout, then gave Codex a self-contained prompt for each phase.
Codex scaffolded authentication, built the server, added tools, wrote and ran tests, fixed failures and deployed to Cloudflare Workers. After each phase, it generated a progress report. De Guzman carried that report back to Claude, checked what had shipped and confirmed completion before proceeding with the next prompt. The author describes ten phases completed in one day; that timing remains his own report.
The result was a 16-tool server for creating, updating and managing articles, music releases, video posts, images and tags. Some compound publishing operations included rollback on failure, while hard deletion required confirmation. The article describing the project was itself published through the pipeline and later edited through its tools.
This account distinguishes the coding agent from the clients that subsequently call the server.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Blog · Ray de Guzman
Frequently asked questions
How did the two coding roles differ?
Claude in Notion planned and documented the phases. Codex implemented, tested, fixed failures and deployed the server tools.
Where did the human intervene?
De Guzman passed prompts and reports between systems, checked each phase and confirmed completion before the next one.
Was the server used for real publication?
The author says the project article was published and later edited through the new MCP tools; hard deletes were confirmation-gated.
More details
Remaining steps
- Author relayed reports and confirmed phases.
About this account
Codex implemented, tested, corrected and deployed self-contained publishing-server phases, returning progress reports for human acceptance.
Primary first-person account opened and checked on 2026-10-14. Bug-resolution and one-day timing are first-person claims, not independent measurements. Outcomes and test claims are attributed to the source and were not independently reproduced. Prompt is suggested by Agents In Use, not the author's quoted prompt.
Credit: Ray de Guzman on Blog
Source published
Added to Agents In Use
Source checked
Tools and topics
Codex · Claude · Notion · Shopify · Cloudflare Workers · MCP
Publishing tools · MCP · Phase review
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.