Use one skill in Claude, ChatGPT, and Codex

Tutorial · 7 min read

If you use more than one AI agent — Claude for one project, ChatGPT for another, Codex for a CLI workflow — you've probably already written the same instructions twice: once as a Claude "skill" or project doc, once pasted into a ChatGPT custom instruction, maybe again as a comment in a Codex config. Every time you improve it in one place, the others quietly go stale.

Skilnu skips the copy-paste by putting the skill in one place and serving it to whichever client asks for it, over the same protocol (MCP) that all three already speak. Connect once per client, and every client reads the current version.

What you need

• A free Skilnu account.
• One skill — write your own, or use one from the catalog.
• An MCP-compatible client. Claude, ChatGPT, and Codex all qualify; so does any other agent tool that supports adding a custom MCP server.

1. Get your MCP server URL

Every Skilnu account has one MCP endpoint. You'll find it in the app under MCP server url, or on your account page — it looks like https://app.skilnu.com/mcp. Copy it; you'll paste the same URL into every client.

2. Add Skilnu as an MCP server

The exact menu differs per client, but the shape is identical everywhere: find the setting for adding a custom MCP server or connector, paste the URL, and authorize with your Skilnu account when prompted.

• Claude — add it as a custom connector/MCP server in your client's connector settings.
• ChatGPT — add it under the client's connector or MCP server settings, the same way you'd add any other tool.
• Codex — register it as an MCP server in your Codex configuration.

Do this once per client. After authorizing, the client can list and load any skill your account can see — your own private skills, anything shared with you, and public skills in the catalog.

3. Load the skill

In each client, ask it to use the skill by name (or let it discover the skill on its own, if the client supports that). The agent fetches the current SKILL.md over MCP and follows it — same instructions, same behavior, regardless of which agent is running it.

4. Update it once

When the process the skill describes changes, edit the skill in Skilnu and publish a new version. You don't touch Claude, ChatGPT, or Codex individually — each one picks up the new version the next time it loads the skill, because they were never storing a copy of it to begin with.

The skill lives in exactly one place. Every client is a reader, not a copy.

Next step

Once a skill is working the same way everywhere, the next thing worth doing is not writing your next skill from scratch — see Compose and version skills like code for how to build on skills that already exist instead.