Use Training Companion with ChatGPT

ChatGPT can use Échapper’s full-access endpoint to read context and propose changes. Review and explicitly approve each exact change before execution. The optional read-only endpoint remains available.

Availability comes first.

Custom MCP apps are used on ChatGPT web, not the mobile apps. Write tools require Developer mode and the relevant account and workspace permissions. Availability and menu names can change.

If Developer mode or Create app is missing, check your plan and ask your workspace administrator before troubleshooting Échapper.

Before you start.

• Create or sign in to your Échapper account.
• Connect Intervals.icu from Échapper Account.
• Use ChatGPT on the web.
• Make sure developer mode and custom app creation are available to your account.

1. Copy the full-access address.

Open AI clients in Training Companion. Copy the address labelled Full access.

Use the address ending in /api/mcp. Existing connections to /mcp or /api/mcp/read remain read-only; update or recreate the app and refresh its tools to opt into controlled writes.

2. Enable developer mode.

Open Settings → Security and login → Developer mode and enable developer mode. Managed workspaces may require an administrator to allow developer mode or app creation first.

3. Create the app.

1. Open ChatGPT Plugins and use the plus button to create a developer-mode app.
2. Name the app Échapper.
3. Paste the full-access MCP address and choose OAuth authentication.
4. Select the authentication option offered by ChatGPT.
5. Choose Scan tools and complete the Échapper sign-in window.
6. Wait for tool discovery to finish, then choose Create.

4. Enable Échapper in a new chat.

The app should appear under Draft apps or Enabled apps with a Dev label. Open a new web conversation and select Échapper from the tools or apps menu.

5. Verify the boundary.

Summarise my rider context, readiness and upcoming calendar. Separate source data from your interpretation.

ChatGPT should be able to retrieve and explain relevant context. Full access also exposes write tools. Request a dry-run preview before any calendar, comment or feedback change. Only execute the unchanged payload with its approval reference after explicit approval. Keep per-action confirmations enabled; do not remember a blanket approval. Read the source afterward to verify the result.

Start with read-only checks.

• Review recent activities and important intervals.
• Compare recovery information with recent training load.
• Read the current calendar and athlete settings.
• Use approved goals, constraints and preferences in an answer.
• Discuss a proposed plan without saving it to the source.

If it does not work.

• No developer mode: check the plan and workspace permissions.
• Scan tools fails: confirm that you pasted the full-access address exactly and that Intervals.icu is connected.
• Authorization closes: repeat the connection and finish Échapper sign-in before returning to ChatGPT.
• App is unavailable in chat: use ChatGPT web and enable the app in a new conversation.
• Tools look old: refresh or recreate the app so ChatGPT scans the current tool list.

For current plan, admin and interface details, consult OpenAI’s developer mode and MCP apps guide.