Connecting real apps to AI agents with the Model Context Protocol
CVolve and Plate Plan include an MCP server, so Claude Desktop and Cursor can work with them. What MCP is, how small a server can be, and our safety rules.
A weekly meal planner with drag-and-drop, smart grocery lists and an AI assistant that plans and evaluates your week, using the AI provider you choose.
Planning meals for a week sounds simple, until you are standing in the supermarket trying to remember what you needed for Thursday. Plate Plan puts the whole week on one board, turns it into a shopping list, and lets an AI assistant do the planning when you do not have the time.
The weekly board shows breakfast, lunch, dinner and snacks for every day from Monday to Sunday. Planning is visual:
Your meal library holds your recipes. When you add ingredients, autocomplete suggests ones you have used before, and every quantity has an explicit unit: 200g, 1 pc, 15 ml, 1 tbsp.
Because every ingredient has a unit, Plate Plan can do the maths for you. The grocery list shows what you need day by day, or as consolidated totals for the week. Three recipes that each use 100g of rice become one line: 300g rice. No more buying the same thing twice.
Open the AI Meal Assistant and ask in plain language:
The planner board and the meal library refresh instantly while the AI works. The chat sits in a compact bar at the bottom of the screen, so your week stays visible the whole time.
Click Evaluate Week and the AI reviews your plan. You get:
Plate Plan always reminds you that AI can make mistakes. For medical or dietary advice, talk to a healthcare or nutrition professional.
Plate Plan is built on Neuron AI, so you can choose your AI provider with one line of configuration: Google Gemini (the default), OpenAI, Anthropic Claude, Mistral, DeepSeek, Grok, or any OpenAI-compatible server.
Want to keep everything on your own computer? Use a local Ollama model. No API key and no data leaves your machine.
And if you do not configure any AI provider, Plate Plan still works. It falls back to local planning and evaluation rules, so the planner, library and grocery list are always usable.
Plate Plan includes a Model Context Protocol (MCP) server and a REST API. Connect Claude Desktop, Cursor or any MCP client, and your agent can read your recipes, plan the week, clear days, build the grocery list and run the nutrition evaluation. That makes Plate Plan a small, practical example of how MCP lets AI agents work with real applications.
git clone https://github.com/future-site-ai/plate_plan.git && cd plate_plan cp .env.example .env.local # optional: add your AI provider and key docker compose up -d
Then open http://localhost:1111. The SQLite database is created automatically and filled with sample recipes.
Plate Plan is deliberately lightweight: plain PHP 8.2, SQLite and vanilla JavaScript with SortableJS for drag-and-drop. There is no heavy framework, which makes it an easy codebase to read, learn from and extend.
Plate Plan is free and open source under the MIT License.
Contributions are welcome: new recipes, AI provider setups, documentation, bug fixes and features. A first automated test suite would be especially valuable. Read the contributing guide and look for issues labelled good first issue.
If Plate Plan helps you plan better meals, please star it on GitHub.
CVolve and Plate Plan include an MCP server, so Claude Desktop and Cursor can work with them. What MCP is, how small a server can be, and our safety rules.