AI and MCP

Ask Claude or ChatGPT about your training.

Connect DataFrame to your AI assistant. It reads the same plan, sessions, and recovery you already sync - you choose what it may view or change, and you can switch that off any time.

AI chat suggesting a HYROX session using DataFrame data

What you can ask

Suggest a session

Replace today's run when legs are tired, keep the plan honest, and send structure to Garmin when you allow it.

Review a workout

Ask how the session went - pace, heart rate, stations, and what faded late.

Compare quality

The same intervals compared with last week - not only the full-run average.

Plan the week

Fill gaps, soften hard days after low readiness, or catch a missed bike without overdoing it.

Example - suggest a session

You ask for a replacement. It builds one and can push it to Garmin.

Real chat with DataFrame connected. The assistant used your plan, recovery, and home-gym kit - then wrote a HYROX session that protects a Saturday long run.

  • You: "can you suggest a session for today - compromised legs as well to replace today's run - also yesterday I did not do the bike"
  • It: Hyrox compromised legs, 6 rounds, about 75 minutes - warm-up, treadmill 1 km + station + rest, optional Z2 bike flush
  • Why: same intent as the replaced run, lighter lunges, stations rotated from last week, short Z2 instead of making up a full missed bike
Open Claude directory
Chat prompt asking for a compromised-legs session and the AI reply with structure

What that reply included

Stations in the example: row 250 m, rope pulls, wall balls, sandbag lunges (20 kg), hard bike, burpee broad jumps plus farmers carry. The assistant explained the load choices in plain words, then noted the missed bike should stay a short Z2 flush - not a full make-up - before a rest day.

When you grant Create or update access, structured work can land on Garmin so the session is on the watch before you leave. View-only is enough if you only want answers in chat.

Example - review a session

You ask how it went. It reads the rounds, not only the average.

After the HYROX session synced, the same assistant answered with a round table, heart-rate drift, and a clear call on what was actually easy.

  • You: "how did the session go?"
  • It: 1 km runs held 4:30-4:31 across six rounds; run HR rose from about 137 to 161 bpm while pace stayed even
  • Aside: the easy bike flush was closer to steady tempo (avg HR 146) - useful before a Saturday long run when daily load was already high
See quality sessions
Chat asking how the session went with a round-by-round AI analysis

Prompts worth trying

Open the prompt in ChatGPT or Claude - then connect DataFrame so the answer uses your real plan and sessions.

  • Suggest

    Using my DataFrame.fit MCP connection, suggest an indoor session for today that keeps intensity but spares my calves.

  • Review

    Using my DataFrame.fit MCP connection, how did yesterday's 14 x 400 m compare to the last time I ran that set?

  • Plan

    Using my DataFrame.fit MCP connection, tomorrow looks hard and readiness is low - what would you soften?

  • Race

    Using my DataFrame.fit MCP connection, I have an Olympic triathlon in nine days - is this week still building or should I start taper talk?

You need DataFrame connected in that assistant for live answers. Without it, the chat still opens with the prompt ready to send.

Assistants DataFrame connects to

Same connection - add one or more, each with its own permissions.

Permissions you control

Authorise screen with View and Create or update columns. Viewing is selected. Changes stay off.

1

Sign in on DataFrame

OAuth so the assistant never sees your Google password.

2

Choose access

View or change - per assistant, not all-or-nothing forever.

3

Ask in chat

Suggest sessions, review quality, compare weeks - using your synced data.

4

Revoke

Disconnect any assistant when you are done.

Claude lists DataFrame.fit in its connectors directory. ChatGPT needs Developer mode and an MCP app URL on dataframe.fit. Cursor has a one-click install. Copilot, Codex, Gemini, Grok, DeepSeek, and other remote MCP clients use https://dataframe.fit/api/mcp. Exact setup steps stay on the in-app help screens after sign-in.

AI and your data

What can the assistant see?

Only what you allow on the authorize screen for that assistant - for example activities, plan, or recovery. It does not get your Google password.

Can it change my plan or push to Garmin?

Only if you turn on Create or update for that assistant. View-only still answers questions from your synced data without writing back.

Does every assistant get the same access?

No. Claude, ChatGPT, Cursor, and others each have their own connection and permissions. Revoke one without touching the others.

Is this a replacement for Home and session quality?

No. Home and activity boards stay the place you decide. The assistant is optional chat on top of the same data.

See your training clearly by tonight.

Get started free