Skip to main content
Whisfit
Open the app (opens in a new tab)
All guides

Connect ChatGPT, Claude, or Gemini

Give an AI assistant read access to your training and nutrition data, and let it write and edit your plans for you.

Time
5 min
You'll need:
A Whisfit account and a ChatGPT, Claude, or Gemini account

Whisfit runs a Model Context Protocol (MCP) server, so an AI assistant can act as your coach: it reads your day, your plans, and your progress, and it can build and edit nutrition and workout plans on your behalf. You ask "am I hitting protein this week?" or "swap Wednesday's chest day for something shorter" and it works with your real data.

Nothing is copied anywhere. The assistant talks to Whisfit directly with your own account, and every request is limited by the same permissions the app uses.

What it can and cannot do

The assistant manages, coaches and can log food for you; workouts and check-ins stay in the app.

It can:

  • Read everything - today's log, your plans, progress, check-ins, workout history, settings.
  • Create, edit, activate, archive, and delete nutrition plans - meals, items, quantities, alternatives.
  • Create and edit workout plans - days, exercises, target sets and reps, supersets, templates.
  • Add and edit your custom foods and custom exercises.
  • Update your settings and targets.
  • Log food to your own day, change or remove a logged item, and mark a meal complete.

It cannot:

  • Log a set, start or finish a workout, or record a check-in.
  • Write to a client's day if you are a coach: it can read what they ate, but only they log it.

Ask it to show you what it logged; it confirms each food and quantity, and logging the same thing twice by mistake never creates a duplicate.

1. Copy the server address

The Whisfit MCP server lives at:

https://mcp.whisfit.com/mcp

That single address works for all three assistants. Keep the /mcp at the end - without it the connection fails.

2. Add it to your assistant

Claude

Available on every plan, including Free (Free allows one connector).

  1. Open Settings → Connectors.
  2. Choose Add custom connector.
  3. Paste the address and select Connect.
  4. A browser window opens with the Whisfit consent screen. Sign in if you aren't already, then approve.

In Claude Code, the same thing from a terminal:

claude mcp add --transport http whisfit https://mcp.whisfit.com/mcp

Then run /mcp to sign in and check the connection.

ChatGPT

  1. On chatgpt.com, open Plugins from the left sidebar (or go to chatgpt.com/plugins).
  2. At the top right, choose Add, then Add custom MCP server.
  3. Fill in the form:
    • Name: Whisfit. The description is optional.
    • Connection: keep Server URL (the other option is Tunnel) and paste the address.
    • Authentication: OAuth, which is the default. Leave "Advanced OAuth settings" alone.
    • The icon is optional (PNG, at least 256x256, max 10 KB).
  4. Read the risk notice, tick I understand and want to continue, then choose Create as a plugin.
  5. ChatGPT opens the Whisfit consent screen. Approve it.

To manage it later, open Settings → Plugins → Whisfit. It shows the connected account, a Permission setting per plugin (Allow low-risk tools by default), and a Refresh tools button. Press Refresh tools whenever Whisfit adds new abilities, such as food logging, so ChatGPT picks them up.

Gemini

Custom MCP servers are supported through Gemini Spark: Settings & help → Connected Apps → Custom apps for Spark, then paste the address.

3. Approve the connection

Whichever assistant you use, the last step is the same: a Whisfit consent screen in your browser, listing which app is asking for access and where it will send you back to.

Read that screen rather than clicking through it. It shows the requesting app's address on purpose, because anyone can build a client that calls itself Claude. If the name and the address don't match what you're expecting, deny it.

4. Try it

Good first questions, once connected:

  • "What did I eat yesterday, and how far off my protein target was I?"
  • "Look at my last month of training and tell me which lifts have stalled."
  • "Build me a 4-day upper/lower plan around the equipment I already use."
  • "My current meal plan is too heavy at dinner - rebalance it across the day, same daily totals."
  • "Log 50 g bagel to my breakfast."

Ask it to explain a change before it makes one. Plan edits are real edits, and the assistant will make them when you ask.

Security and privacy

  • Access is granted with OAuth, the same standard "sign in with" flow you already use elsewhere. The assistant never sees your password.
  • The assistant acts strictly as you. It cannot reach another user's data, because the database enforces that per-row, not the assistant.
  • Revoke any time - remove the connector in your assistant, and its access ends.
  • Anything you ask about goes through your assistant provider's servers, exactly as any other chat message does. Their privacy policy governs the conversation; ours governs the data in Whisfit.

When it goes wrong

  • "Not authorized" after signing in successfully. You approved against a different environment than the connector points at. Check the address is the one above, with no trailing characters.
  • The connector shows no tools. The connection was added but not authorized - reopen it and complete the consent step.
  • Gemini shows no "Custom apps" option. You're outside the US eligibility window described above.
  • The assistant says it can't log food. Its tool list is out of date: open Settings > Plugins > Whisfit and press Refresh tools.
  • It gets "today" wrong. Tell it the date. The server never assumes what day it is for you - it runs in UTC, and guessing your local day would quietly write to the wrong one.