Mcprunbook

OpenAI renamed ChatGPT apps to plugins, the third rename in three years. No redirects, no status page for connectors, and when their pipeline fails, it fails in our name.

How endurance coaches use AI assistants like Claude and Perplexity with the Tredict MCP Server to build a reusable training plan library that syncs structured workouts to Garmin, Coros and other watches.

Perplexity as MCP client connects to external tools. See how it works with Tredict to analyse running data, determine FTP, and answer training questions with real-time web context.

Raw ground contact time numbers are misleading because GCT drops at higher speeds. Claude calculated a normalized efficiency value from my actual training data via the Tredict MCP Server.

A working example of Interactive MCP Apps, also known as ChatGPT Apps on the OpenAI side. Tredict ships two MCP tools that render actual widgets inside Claude.ai and ChatGPT, not plain text summaries.

Claude.ai is great for day-to-day training questions. Claude Code in the terminal is built for long-running tasks, which turns out to matter a lot when you want a full season plan written directly into Tredict.

Same data, same prompt, three AI assistants. Why ChatGPT and Mistral fail at training plans and what makes the difference.

How to connect the Tredict MCP Server to Claude, ChatGPT, Codex and Mistral. Which client works best and why.

A technical look at what data and actions Claude or ChatGPT can work with through the Tredict MCP Server, and why that changes what AI can do with training data.

After two colds in six weeks I used Claude and the Tredict MCP Server to get back into running. I was sceptical. After two weeks I was not.