Recipe Manager MCP Server

A FastMCP server built with FastAPI that serves a recipe editing web page and exposes recipe data via MCP tools. Designed for voice-assistant interaction — fractions like 1/2 and 1¼ are automatically normalized to "a half" and "one and a quarter" so they read naturally aloud.

Features

• Web UI — Clean, responsive form with Edit and Import tabs at /
• Recipe Import — Paste plain-text recipe blocks and auto-parse into structured fields
• Fraction Normalization — 1/2, 1¼, ½ → "a half", "one and a quarter", "a half" on save
• Notes Field — Optional notes section for tips, substitutions, and comments
• REST API — GET /api/get-recipe, POST /api/save-recipe, POST /api/import-recipe
• MCP Tools — get_recipe, save_recipe, and import_recipe exposed at /mcp/
• Persistent storage — Recipes saved to recipes.json in a Docker named volume

Quick Start

Running with Docker (recommended)

bash run.sh

This rebuilds the image and starts the container on port 8002 with a named volume (recipe-data) for persistent storage and --restart unless-stopped for auto-recovery.

Running locally

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn server:app --host 0.0.0.0 --port 8000

Web UI

Open http://localhost:8002 in your browser.

Edit Tab

Four fields: Recipe Name, Ingredients, Directions, and Notes. Edit any field and click Save Recipe to persist changes. Fractions are normalized on save so voice assistants read them naturally.

Import Tab

Paste a plain-text recipe block and click Parse & Save. The parser extracts the title, ingredients, directions, and notes, then switches to the Edit tab so you can review before saving.

Expected format:

Recipe Name Here

Ingredients:
1 1/2 cups flour
1/4 tsp salt
2 eggs

Directions:
Mix ingredients together.
Bake at 350°F for 25 minutes.

Notes:
Add vanilla extract for extra flavor.

The first line becomes the recipe title. Sections are identified by Ingredients:, Directions:, and Notes: headers (case-insensitive). Nutrition facts and source URLs are ignored.

Endpoints

Endpoint	Method	Description
/	GET	Web UI for editing and importing recipes
/api/get-recipe	GET	Get the current recipe as JSON
/api/save-recipe	POST	Save a recipe (JSON body with name, ingredients, directions, notes)
/api/import-recipe	POST	Import a plain-text recipe (JSON body with text)
/mcp/	POST	MCP Streamable HTTP transport endpoint

MCP Tools

• get_recipe — Retrieve the current recipe
• save_recipe — Save or update a recipe (params: name, ingredients, directions, notes)
• import_recipe — Parse and save a plain-text recipe block (param: text)

Import via MCP example

import_recipe(text="Classic Pancakes\n\nIngredients:\n1 1/2 cups flour\n1/4 tsp salt\n\nDirections:\nMix and cook on a griddle.\n\nNotes:\nServe with maple syrup.")

Fraction Normalization

On save, fractions in ingredients, directions, and notes are converted to voice-friendly words:

Input	Output
1/2 cup	a half cup
1 1/2 cups	one and a half cups
3/4 tsp	three quarters tsp
1¼ cups	one and a quarter cups
½ tsp	a half tsp
80/20 ground beef	80/20 ground beef (unchanged)

This ensures voice assistants read measurements naturally instead of saying "one slash two."

Project Structure

recipe-mcp/
├── Dockerfile            # Container build definition
├── .dockerignore         # Docker build context exclusions
├── .gitignore            # Git exclusion rules
├── README.md             # This file
├── data.py               # Recipe I/O, Pydantic model, and import helper
├── mcp_server.py         # FastMCP tools definition
├── parser.py             # Plain-text recipe parser and fraction normalizer
├── requirements.txt      # Python dependencies
├── run.sh                # One-command rebuild + restart (port 8002)
├── sample_1.txt          # Sample recipe for testing import
├── sample_2.txt          # Sample recipe for testing import
├── server.py             # FastAPI app, routes, and entry point
└── templates/
    └── web_page.html     # Web UI template with Edit/Import tabs

Recipe data is persisted in a Docker named volume (recipe-data) and is not tracked by git.