How to use the same skills in Claude and ChatGPT

Why the two drift apart

A skill is a folder with a SKILL.md file: a short YAML block with a name and a description, then the instructions. Claude takes that folder as a ZIP, and ChatGPT's Skills upload from your computer too. The instructions move between them without rewriting.

The copies don't. Claude keeps the skills you upload in your Claude account; ChatGPT keeps its own in your ChatGPT workspace. The day you fix a step in Claude, ChatGPT still runs the old one, and so does every teammate who uploaded it last month.

Plans add a second gap. Claude Skills work on every plan, Free included. ChatGPT Skills exist only on Business, Enterprise, Healthcare, and Edu, so a Plus or Pro user has nowhere to upload the file at all.

The do-it-yourself way: one folder, two uploads

1. Keep the master copy in one place, such as a shared drive folder or a Git repo, one subfolder per skill.
2. Zip and upload to Claude: Customize > Skills > "+" > "+ Create skill" > "Upload a skill".
3. Upload to ChatGPT on Business or above ("Upload from your computer" in Skills). On Plus or Pro, paste the instructions into a Project instead.
4. On every change, edit the master copy and upload again in both products, for every person.

For one person with three skills, this is fine. For five people and twenty skills, it is a hundred uploads per round of edits.

The connected way: both products read one server

Claude and ChatGPT both connect to remote MCP servers. Serve the skills from one server and each AI reads the current version when it needs it.

• Claude: Customize > Connectors > "+ Add" > "Add custom connector". On every plan; Free allows one.
• ChatGPT: Plugins > "+" > "Add custom MCP server". Full MCP on Business and Enterprise/Edu, read/fetch on Pro; Plus only where the plan allows custom connectors.

Two things change when you serve skills this way. Claude chat and Desktop don't list MCP-served skills in their Skills panel; the AI reaches them through the connector. And because both read the same server, there is nothing to re-upload.

Write skills that behave the same in both

• Keep the description concrete. Both products pick a skill by its description. "Drafts refund replies in our support tone. Use when a customer asks for a refund" works in either.
• Prefer instructions over scripts. Plain steps run the same everywhere. A bundled script depends on each product's code execution.
• Avoid product names in the steps. "Ask the user for the order number" beats "use Claude's file tool".
• Test once in each. Ask the same question in Claude and ChatGPT and compare.

The coding agents read the same format from folders on disk; the sync guide lists where each one looks.

Questions

Is there a way to share one set of skills across Claude, ChatGPT and Cursor?

Yes: keep the skills on one MCP server and connect each tool to it. Claude (every plan), ChatGPT (Pro, Business, Enterprise; Plus where custom connectors are allowed), and Cursor all accept a remote MCP server.

Does a skill written for Claude work in ChatGPT?

Usually, if it is plain instructions. Both use the SKILL.md format with name and description frontmatter. A skill that runs bundled scripts depends on each product's code execution, so test it in both.