Training Tilt MCP Server

A quick note before you start: the MCP server is not an AI coach. It doesn't replace you, and it isn't trying to. It's there to help you do what you already do, faster and easier, by giving Claude or ChatGPT access to your Training Tilt data.

Claude and ChatGPT are both really good at reading, writing, and reasoning, but on their own, they only know what's in the conversation. They can't see your Training Tilt data unless you copy and paste it in, or go find it yourself by clicking around the app.

An MCP server fixes that. It's a piece of software that sits between an AI and a specific app, in this case Training Tilt, and lets the AI ask for exactly the data it needs, or make a specific change, without you doing the copy-pasting or the navigating yourself. Think of it as a translator: the AI says "get me this athlete's last month of training," the MCP server turns that into a proper request to Training Tilt, and hands the answer back.

MCP itself is an open standard. The same idea works whether you're using Claude, ChatGPT, or any other AI that supports it, not something unique to one company. But the server itself isn't automatic just because the standard is open: Training Tilt is the one that had to build it, and is the one that keeps it running, maintains it, and supports it. Once it's connected, the AI can use it in a conversation the same way it uses its own knowledge, just pointed at your real coaching data.

The Training Tilt MCP server doesn't run any AI itself. It's a plain data connector, it fetches and saves the specific things Claude or ChatGPT asks for (an athlete's training log, a planned workout, a comment) and hands that data back and forth.

All the actual thinking, reading your prompt, deciding what to pull, writing a workout, drafting a comment, comes from Claude or ChatGPT, not from Training Tilt. The MCP server is just the pipe between the AI and your data.

That also means behavior can vary a bit depending on which AI you're using and how you've prompted it. The examples below describe what should happen, but the AI is making its own call in the moment based on your prompt, so it's worth checking the result before it goes to an athlete.

Where you draw the line between "the AI pulls data" and "the AI does the coaching" is up to you, that's a conversation between you and Claude or ChatGPT, not something the connector decides. Some coaches will use it mainly to read and summarize: pulling a season's training load in seconds instead of clicking through the app. Others will go further and build custom workflows, teaching the AI their own coaching principles, feedback style, and periodization approach, so it drafts workouts or check-ins that already sound like them. Skills, projects, and scheduled tasks are what make that repeatable: save your approach once, and the AI keeps applying it automatically instead of you re-explaining it every session.

What the AI is genuinely good at is reading, distilling, and summarizing large amounts of data, and producing a fast first draft from your instructions. It's not good at replacing the coaching itself, the judgment calls, the relationship, the read on how an athlete is actually doing. That part stays with you. What changes is how much time goes into clicking around the Training Tilt app versus actually thinking and coaching.

Try these first (once connected, see setup steps at the end):

That's enough to get moving. The rest of this guide is a reference for when you want to do more, with setup steps at the end.

Anything sent to an athlete, a comment, a note, a scheduled workout, should be shown to you first. Claude or ChatGPT shouldn't post or schedule anything without your OK, unless you tell it to skip the review (e.g. "just send it").

To act on a specific athlete, the AI needs their athleteId. It should get this from a roster search, you shouldn't need to know or supply it yourself, just say the athlete's name.

This is just a reference guide and will change over time as we learn and extend the server. You should regularly "refresh" the tools in ChatGPT or Claude to make sure the AI has the most up to date reference.

The best way to find out what the MCP server is capable is to ask Claude or ChatGPT. What can you help with with from the "Training Tilt" connector.

Two paths, and the AI should ask which one fits if it's not obvious from your prompt:

For structured sessions (intervals, specific paces or power targets), the AI should ask whether you prescribe by zone number or percent of threshold, it shouldn't look up an athlete's actual zones or FTP, just build the workout using whichever style you tell it to use.

Strength sessions are the exception, these should stay as plain sets/reps in the description rather than stepped structure, unless you specifically ask for something like a structured erg or timed circuit.

Once you're comfortable with the basics, these are the kinds of things worth building toward.

Instead of pulling up the athlete's page yourself, hand the AI the context ahead of time and let it do the reading:

"I've got a call with [athlete] coming up. Give me their training load trend over the last month, how consistent they've been with the plan, anything they've flagged in comments, and where they're at against their goal race. Highlight anything else you think I should know from the data."

One prompt replaces several tabs of clicking, and gives you time to actually think about what you want to cover, not just skim it right before you dial in. That last line is worth keeping as a habit across prompts generally, it gives the AI room to surface something you didn't think to ask about.

Between comments on workouts and notes threads, it's easy for something to slip through, especially across a full roster. Instead of clicking into each athlete's thread one by one:

"Pull my most recent conversations across all athletes, comments and notes both, and flag anything I haven't replied to or might have missed."

This is a good one to run first thing in the morning or whenever you're catching up after being away, it turns a scroll through every athlete's page into a single list of what actually needs your attention.

This is where scheduled tasks earn their keep. Rather than opening the app and clicking into each athlete one by one, you can have a digest waiting for you every Monday:

"Every Monday at 7am, go through my whole roster and pull: last week's completed activities, this week's planned workouts, and any recent comments or notes. Flag anyone who missed more than one planned session, anyone whose training load jumped sharply, and anyone with a comment or note that still needs a reply from me. Give me a short summary I can scan before my day starts."

The AI does the roster-wide reading and distilling, completed sessions, what's coming up, and the conversation threads, so nothing sits unanswered just because it scrolled off your screen. You spend your time on the athletes who actually need something from you that week, not clicking into everyone to check.

None of these need to be exact, they're starting points. The more specific you are about how you coach (your onboarding style, what counts as a red flag, how you phrase feedback), the more the AI's output starts to sound like something you'd actually send yourself.

A few things worth knowing:

You'll get a Client ID and Client Secret from Training Tilt, keep the secret private, treat it like a password. Both go into the connector setup below, one time, on whichever platform you use.

Available on Free, Pro, Max, Team, and Enterprise plans, Free accounts are limited to one custom connector. On Team/Enterprise, an Owner needs to add the connector at the org level first before individual coaches can connect.

Developer Mode is available on Pro, Plus, Business, Enterprise, and Education web accounts, and supports both read and write actions, scheduling workouts and leaving feedback should work the same as on Claude. ChatGPT will ask for confirmation before running write actions by default, same principle as the review step described above. It's still labeled beta by OpenAI, so worth testing the same way you tested the Claude side before relying on it for real athlete work.

Training Tilt adds and improves tools on the connector over time. Every now and then, especially if something you'd expect to work seems missing, it's worth refreshing the connection so you're on the latest version:

You don't need to do this constantly, just when you hear about new features or notice something isn't behaving the way this guide describes.