Use BetterMenu with an AI assistant

You want your AI assistant to help you build a recipe, search for the right ingredients, and calculate a nutrition label — without switching between your AI tool and BetterMenu. BetterMenu connects directly to Claude, Claude Code, and other compatible AI assistants, giving them access to your full recipe library and every BetterMenu tool. Every change happens in real time against the same record — ask from your phone or work in Studio at your desk, and BetterMenu stays the single source of truth.

How does the connection work?

BetterMenu provides a server that speaks the Model Context Protocol (MCP) — a standard way for AI assistants to call external tools directly. Once your AI tool is connected, it can use BetterMenu tools as naturally as it uses any other capability: searching the USDA ingredient database, creating and updating recipes, computing nutrition facts, configuring serving sizes, and exporting labels.

The AI assistant acts on your behalf, using the same permissions you have in BetterMenu. It can read and create recipes within your organization, but it cannot access data belonging to other organizations. An R&D team using Claude to prototype new granola bar formulas, for example, sees only the recipes that team already has access to in BetterMenu Studio — the connection does not grant broader access. The result is that your AI assistant becomes a hands-free interface to BetterMenu: you describe what you need, and it handles each step for you.

Does the AI assistant ever guess at the nutrition numbers?

No. BetterMenu's AI assistant integration is there for the busywork — parsing your plain-language request, searching the USDA database, drafting the recipe, figuring out what to ask you next. It never estimates a nutrition value itself. Every number on the label comes from BetterMenu's deterministic calculation engine: the same USDA FoodData Central lookups and 21 CFR §101.9 rounding rules used when you build a recipe manually in Studio. Your AI assistant calls that engine as a tool — it doesn't predict, interpolate, or guess at a single gram, milligram, or percentage.

This isn't a matter of trusting the underlying AI model to behave — it's enforced by how BetterMenu's AI assistant integration is built. None of its MCP tools accept or return a free-form nutrition value from the model; each tool call runs the same calculation engine and hands back a number the engine computed, with the USDA record and rounding rule that produced it. The model has no path to inject a guessed number even if it wanted to, and that guarantee comes from BetterMenu's tool design, not from the model's behavior — it holds no matter which AI assistant, Claude, Copilot, or any other MCP-compatible tool, is on the other end of the connection. Because the engine reasons the same way whether a human or BetterMenu's AI assistant integration triggered it, you get the same detailed calculation trail either way: which USDA record was matched, and which rounding rule produced each value.

This distinction matters because general-purpose AI is measurably unreliable at estimating nutrition data on its own. A 2025 study in Nutrients comparing professional dietitians against five AI chatbots on real packaged-meal nutrition labels found dietitians accurate to within 88–99.5% across calories, protein, fat, and sodium — while the chatbots' unaided estimates ranged wildly, from a 39% carbohydrate underestimate to a 200% fat overestimate. Sodium was "consistently underestimated across all AI models," with the researchers concluding AI output is fit only for preliminary assessment, not compliance-grade use (Hsuan et al., Nutrients 2025). BetterMenu's AI assistant integration sidesteps that failure mode entirely by never letting the model estimate in the first place — it's the deterministic engine doing the math, every time.

Is a nutrition label created through an AI assistant FDA-compliant?

Yes — because compliance comes from BetterMenu's calculation engine, not from whichever AI model your assistant runs on. Whether you build a recipe by hand in Studio or ask an AI assistant to do it, the label is generated by the same USDA-sourced, 21 CFR §101.9-compliant pipeline described in Create a nutrition label. BetterMenu's AI assistant integration is an interface to that pipeline, not a substitute for it — the model never generates a compliance number on its own. Every action it takes is also attributed to the authorizing account in the audit trail, so AI-assisted changes remain exactly as traceable as manual ones — nothing is attributed to an anonymous agent.

What can my AI assistant do?

A connected AI assistant can run the full BetterMenu recipe workflow without any manual steps. You describe what you want in plain language, and the assistant calls the appropriate BetterMenu tools in the right order.

The table below maps common plain-language requests to the BetterMenu action the AI assistant performs behind the scenes. Each row is a complete operation — the assistant handles the sequencing, so you do not need to know which tool to call or in what order.

Your AI assistant can also query your recipe history, read the audit trail, and retrieve past label exports — all through the same connection, using the same plain-language approach shown above. For nutrition analysis workflows, an AI agent can loop through multiple recipes automatically, comparing formulas and flagging differences without manual effort from your R&D team. The result is faster iteration: your team describes what they need, and BetterMenu acts on it directly.

What access does the AI assistant have?

The AI assistant uses your BetterMenu account's permissions. It can access every recipe and ingredient you have access to within your organization, and it respects the same role restrictions as the BetterMenu Studio interface. If you have Viewer access to a recipe, the AI assistant also has Viewer access and cannot make changes to it. Editor and Admin roles work the same way — the connection does not elevate permissions.

All actions taken by the AI assistant are recorded in the audit trail under your account, with a note indicating the action was performed via an AI tool. This means your recipe history and compliance records remain complete even when an AI assistant is doing the work. For regulatory affairs teams who need a clear record of every formula change — including changes made during AI-assisted R&D workflows — the audit trail captures the full picture. Nothing is attributed to an anonymous agent; every action traces back to the authorizing team member.