How to log food with ChatGPT

Short answer: connect Forkmate to ChatGPT once as a remote MCP server, then log a meal by typing it in plain language — “I had a chicken burrito and a banana for lunch” — and ChatGPT records it to your Forkmate diary. The Forkmate server URL is:

https://mcp.forkmate.ai/

One honest caveat up front: adding your own MCP server to ChatGPT currently needs Developer mode, a beta that provides full (read + write) MCP connector support on ChatGPT’s Plus, Pro, Business, Enterprise, or Edu plans, and is web-only for now. If your ChatGPT can enable Developer mode, Forkmate works. If it can’t yet, Claude and Cursor support the same one URL today — the rest of this guide covers both the ChatGPT path and the fallback.

Why log food by talking to ChatGPT?

Most calorie trackers make you stop what you’re doing, open an app, search a database, and tap through portions. Forkmate flips that: it’s a remote MCP server, so the AI assistant you already talk to can do the logging. You describe the meal the way you’d describe it to a friend, and it lands in a clean calorie-and-macro diary you can open any time.

The one-time cost is connecting ChatGPT to Forkmate. Do that once and logging becomes a sentence in a chat you were already having.

Can ChatGPT actually connect to a custom MCP server?

Honestly, it depends on your plan — so here’s the current state, plainly:

ChatGPT plan	Custom (self-added) MCP servers	Notes
Free	Not yet	No Developer mode; use Claude or Cursor for now
Plus / Pro	Yes, via Developer mode (beta)	Full read + write MCP tools; web client
Business / Enterprise / Edu	Yes, via Developer mode (beta)	An admin may need to enable it

Developer mode is a beta, OpenAI is broadening it over time, and the “connectors” you see in the default ChatGPT UI are OpenAI-maintained wrappers for big services — not the place to paste a third-party server. Adding your own server (like Forkmate) is what Developer mode unlocks. We’ll keep this post updated as that support widens.

Before you start

• A free Forkmate account. If you don’t have one, create your free account first, then come back.
• A ChatGPT plan that supports Developer mode (see the table above), or a Claude / Cursor client as the fallback.
• The Forkmate MCP URL: https://mcp.forkmate.ai/. The same URL works everywhere.

Option A — ChatGPT (Developer mode)

1. In ChatGPT, open Settings and enable Developer mode (in the connectors / apps area; on Business/Enterprise/Edu an admin may need to turn it on first).
2. Add a new MCP server and paste the Forkmate URL: https://mcp.forkmate.ai/
3. When ChatGPT opens the Forkmate sign-in, sign in once to approve access. That secure OAuth step is what connects it to your diary and nobody else’s.
4. Back in the chat, tell it what you ate and it logs through Forkmate’s tools.

That’s the whole setup. There’s no separate app to install or babysit — Forkmate lives behind that one connection, and you can disconnect any time from the same screen.

Option B — the fallback (Claude or Cursor)

If your ChatGPT plan can’t add a custom server yet, the same Forkmate URL works in Claude and Cursor today. Claude Code adds it from the terminal in one line:

claude mcp add --transport http forkmate https://mcp.forkmate.ai/

…or paste the URL into Claude’s Connectors screen, or Cursor’s MCP settings. Full per-client steps are on the connect page.

Log your first meal

Once connected, just tell ChatGPT what you ate:

I had a chicken burrito and a banana for lunch.

It logs the meal through Forkmate — meal, portions, and macros. You’ll see the calories and macros it’s about to save, so you can confirm or correct the numbers before they’re committed. A few more that work well:

• “Add 2 eggs and a slice of sourdough toast to breakfast.”
• “I had a grande latte with oat milk this morning.”
• “Log 200g of grilled salmon and a cup of rice for dinner.”

Then open the web diary for a clear view of your day: a calorie ring, protein / carb / fat bars, and every meal grouped like the trackers you already know.

How accurate are the numbers?

This is the honest part. AI is genuinely good at identifying what you ate from a description, but estimating portions and calories is where any tracker — human, database, or AI — carries real error. Peer-reviewed studies put AI calorie estimates well off on mixed meals, and portions are the hardest part to guess. We wrote up what the research actually says if you want the numbers.

Forkmate’s answer isn’t to claim we’re more accurate than everyone else — it’s to keep you in the loop. It resolves foods against open, public databases (USDA FoodData Central and Open Food Facts) and shows you the calories and macros before they’re saved, so you confirm every number instead of trusting a black box. If a portion looks off, you fix it in the chat or the diary before it’s committed.

Our take: most people quit tracking because it’s tedious, so logging consistently and easily matters far more than chasing perfect numbers — good-enough numbers you actually log beat perfect numbers you abandon. If you need pinpoint accuracy for clinical reasons, weigh your food and enter it precisely in the web diary, and talk to your physician. Forkmate is a general-wellness tool, not medical advice.

Coming from another app?

If you’re moving over from MyFitnessPal, you don’t have to start from zero — you can bring your whole history across for free, and we won’t re-estimate your old numbers.

Not sure what your daily target should be? Get a calorie goal in about 60 seconds before you log a thing.

Recap

1. Enable Developer mode in ChatGPT (Plus/Pro/Business/Enterprise/Edu) and add the Forkmate MCP URL — or use Claude / Cursor if your plan can’t yet.
2. Approve the one-time sign-in.
3. Tell ChatGPT what you ate, confirm the numbers, and check your diary.

That’s it — from here, logging your food is just talking to the assistant you already use.