Blog ·

How to sync agent skills between ChatGPT and Claude

Short answer: ChatGPT and Claude don’t share skills natively — there is no built-in sync between them. The reliable fix is to keep your skills in one place and connect both clients to it over MCP (the Model Context Protocol). Then the same governed, versioned agent skills load in either ChatGPT or Claude, and you update them once.

If you use both assistants — Claude for some work, ChatGPT for other — you have probably hit the wall: a skill you set up in one isn’t there in the other. That’s expected. Each vendor manages its own capabilities and doesn’t export them. Below is why, and the connect-once pattern that gives both assistants the same library.

Why skills don’t sync between them

Two separate gaps cause this:

  • Claude has skills, but they’re siloed. Anthropic’s custom skills are tied to where you set them up and do not automatically follow you across every surface or to other vendors’ products.
  • ChatGPT has no native “skills” library equivalent to Claude’s. So there is nothing on the OpenAI side for Anthropic skills to sync into — and vice versa.

The one thing both do share is the open standard underneath: MCP. ChatGPT, Claude, and most major assistants can connect to an MCP server as a custom connector. That makes MCP the bridge — not a vendor feature, but the neutral layer both speak.

The fix: one skill library, served over MCP

Instead of recreating skills in each assistant, host them once on an MCP skill server and point both ChatGPT and Claude at it. Now “sync” isn’t a copy operation that drifts — both clients read the same source, so they’re always consistent.

Connect Claude

  1. In Claude (claude.ai or Desktop), open Settings → Connectors and Add custom connector.
  2. Point it at your skill library’s MCP endpoint (for Skills Manager: https://www.skillsmanager.uk/mcp) and authorize with OAuth.
  3. For Claude Code / Codex, add the same URL with a scoped bearer token instead of OAuth.

Connect ChatGPT

  1. Open ChatGPT on the website (not the mobile/desktop app).
  2. Go to Apps / Settings and turn on Developer mode.
  3. Choose Create custom connector and use the same MCP URL, then save and connect.

ChatGPT vs Claude skills today

ClaudeChatGPT
Native skills featureYes (custom skills)No equivalent library
Syncs skills to the other vendorNoNo
Connects to an MCP serverYes (connector / OAuth)Yes (developer-mode connector)
Result with a shared MCP librarySame skills, always currentSame skills, always current

Keeping the two genuinely in sync

Connecting both isn’t enough on its own — you want a single canonical version, not two copies you maintain. The durable setup:

  • Git is the source of truth. Skills live in a repo; edits are reviewed and versioned. The hub syncs from Git, so both ChatGPT and Claude always get the current version.
  • One change propagates everywhere. Update a skill once and every connected client — ChatGPT, Claude, Codex, others — sees it on the next session.
  • Access stays governed. Org/team/user permissions and scoped tokens apply no matter which assistant connects.

Make both assistants actually use the skills

Some clients only show a truncated preview of available skills and won’t load the full list unprompted. Add a one-line instruction (in Claude’s or ChatGPT’s custom instructions) telling the assistant to load your favourite/starred skills at the start of every session, so the right skill is always in context. Skills Manager has a ready-made enforce-loading setup for each client.

Bottom line

There is no ChatGPT ↔ Claude skill sync button, and there won’t be — they’re competing products. But you don’t need one. Put your skills behind a single MCP library, with Git as the source of truth, and both assistants read the same governed set. That’s the only “sync” that doesn’t drift.

Related: what are agent skills, why skills, not just MCP, and how to connect your assistants.

One library, every assistant. Start free and connect ChatGPT and Claude to the same governed skills.

Last updated: June 9, 2026

Serve the same skills to every assistant