This is a submission for the Notion MCP Challenge

What I Built

I built Notion Skills Registry, a lightweight package-registry-style system for AI skills powered by Notion MCP.

The problem is simple: once you start using AI coding agents seriously across multiple repos, skill management becomes messy fast.

Prompts get duplicated.

Global skill folders become cluttered.

Different projects end up using different versions of the same workflow.

And once a team starts sharing skills, things turn into copy-paste chaos.

So I built a better pattern:

• keep the real skill content in Notion
• organize skills by workflow or use case
• version them cleanly with names like content-research-writer@v1 and content-research-writer@v2
• publish approved versions through a central Skill Versions page
• install only the exact versions each repo needs into Claude Code or OpenCode

Instead of stuffing everything into one global skills folder, each project gets a clean local wrapper skill, while the source of truth stays centralized in Notion.

That gives me the best of both worlds:

• centralized skill authoring
• per-project isolation
• version control for prompts and workflows
• easier team sharing
• a much cleaner upgrade path

The result feels like a package manager for AI skills, except the registry is a Notion workspace and distribution happens through Notion MCP.

Why this is useful

This unlocks a much more scalable workflow for agentic development.

A single Notion workspace can now act as:

• a skill registry
• a version history for operational prompts
• a distribution layer for AI workflows
• a lightweight governance system for approving what projects are allowed to install

That matters because AI skills are turning into real infrastructure.

Once teams rely on them for research, content ops, analysis, coding workflows, QA, or internal automations, you need more than random markdown files copied across repos.

You need:

• a source of truth
• clean versioning
• controlled rollout
• a way to keep projects in sync without breaking everything

That is exactly what this project is designed to do.

How it works

At a high level:

1. Skills live in Notion as normal pages.
2. Each version gets its own page, such as content-research-writer@v1.
3. A Notion page called Skill Versions acts as the registry index.
4. Management skills handle install, update, and version checks.
5. A repo installs a local wrapper skill that points to the exact Notion page.
6. The real instructions remain centralized in Notion.

The current workflow includes four management commands:

• /notion-init
• /notion-install
• /notion-update
• /notion-check-updates

These commands make it possible to initialize a local registry manifest inside a repo, install a skill from Notion, detect newer approved versions, and update wrappers without manually copying prompt files around.

Example structure

Inside Notion, the registry looks like this conceptually:

[TEMPLATE] Notion Skills Registry/
  Skills/
    Content Research Writer/
      content-research-writer@v1
      content-research-writer@v2
    Video Downloader/
      video-downloader@v1
  Skill Versions/
    Content Research Writer
      content-research-writer@v2 - approved
      content-research-writer@v1 - approved
    Video Downloader
      video-downloader@v1 - approved

That means teams can keep drafts, multiple releases, and approved production-ready versions all in one place.

Then per repo, the installed output stays clean:

.opencode/
  notion-skills.json
  skills/
    content-research-writer/
      SKILL.md

or:

.claude/
  notion-skills.json
  skills/
    content-research-writer/
      SKILL.md

What makes this interesting

A lot of people use Notion as documentation.

This project uses Notion as live skill infrastructure.

That shift is what makes it exciting.

Notion stops being just a place where instructions are stored and starts becoming the operational registry that AI agents can install from.

So instead of asking:

Where is the latest version of this prompt?

You can ask:

Which approved version should this repo install right now?

That is a much more useful primitive for teams building with agents.

Video Demo

Here is a walkthrough of the workflow in action:

Show us the code

GitHub repo:

Notion template:

Notion

A tool that connects everyday work into one space. It gives you and your teams AI tools—search, writing, note-taking—inside an all-in-one, flexible workspace.

skills-mcp.notion.site

How I Used Notion MCP

Notion MCP is the core enabler here.

I used it to turn a Notion workspace into something installable and operational for AI coding workflows.

More specifically, Notion MCP makes it possible for the management skills to:

• access the registry page directly from Notion
• resolve linked skill version pages
• load the exact skill content associated with a selected version
• keep local project wrappers connected to a centralized source of truth

That unlocks a workflow where Notion is no longer just static reference material.

It becomes:

• a registry
• a release surface
• a collaboration layer
• a versioned distribution system for agent skills

In practice, that means I can author and revise skills in Notion, approve the versions I want projects to consume, and let repositories install either:

• the latest approved release
• or a pinned version for stability

That model is powerful because different repos often need different levels of change tolerance.

Some should auto-follow the latest approved skill.

Others should stay pinned until someone intentionally upgrades them.

Notion MCP gives me the bridge between those two worlds: centralized authoring and local execution.

What this unlocks beyond this challenge

I think this pattern has a lot of room to expand.

This could evolve into:

• shared team skill catalogs
• internal skill release workflows
• approval pipelines for prompt changes
• skill packs by department or project
• multi-tool support beyond Claude Code and OpenCode
• analytics around which skills are installed where

In other words, this is not just a Notion page with prompts.

It is the beginning of a real skill distribution system for agentic work.

Why I think this is challenge-worthy

This project is useful immediately, but it also points toward a bigger idea.

As AI agents become part of everyday development workflows, teams will need real systems for managing the instructions those agents rely on.

Notion MCP makes that possible in a surprisingly natural way.

It lets a familiar workspace become programmable infrastructure.

That is why I think this project stands out:

• it solves a real workflow problem
• it uses Notion MCP in a way that is practical, not decorative
• it creates a reusable system others can adopt
• it turns Notion into something much more operational than a notes app

Closing

If you are building with Claude Code, OpenCode, or similar agent workflows, and you are tired of duplicating prompts across repos, this pattern is worth trying.

Notion Skills Registry gives you one place to create, organize, version, approve, and distribute AI skills cleanly.

That is the workflow I wanted for myself, so I built it.

If this resonates, check out the template and adapt it to your own agent stack.

Template: Notion Skills Registry Template