Why ChatGPT is read-only for now

ChatGPT is capable of working with MCP tools, and OpenAI is expanding write support on selected plans. Échapper still exposes only read tools through its ChatGPT address. That is a product boundary, not a claim about what ChatGPT can do in general.

Reading and changing are different products.

Reading training context helps answer questions: what happened, what is planned and what might fit next. Writing changes the source of record. A calendar edit, activity comment or saved rider observation can affect future decisions and deserves a stricter interaction.

The boundary is enforced by the server.

Training Companion does not rely on a warning inside the guide. ChatGPT receives a separate MCP address whose discovered tool list excludes create, update, delete, comment, feedback and profile-saving tools.

Even if a prompt asks ChatGPT to make a change, there is no Échapper write tool available through that address.

Claude uses a different path.

Claude and compatible write-capable clients can use the full-access address. A write request still does not run immediately: Échapper first returns an exact preview, then accepts only the unchanged, short-lived action the rider explicitly approves.

Read access should be easy to understand. Write access should be hard to mistake.

Why keep the restriction now?

• The public Intervals.icu connection is still awaiting approval.
• ChatGPT plan and workspace capabilities continue to change.
• A separate read-only route is easier for riders and reviewers to verify.
• It lets the product learn from real read workflows before expanding another action surface.

What would need to change.

ChatGPT write access will be reconsidered only when the complete experience can preserve the same visible preview, payload binding, explicit approval and audit trail already required by Échapper’s full-access route.

Until then, use the ChatGPT guide for analysis and the Claude guide when you deliberately want controlled write capability.