Food Recipe MCP

What AIDataNordic/Food-Recipe-MCP MCP server does

AIDataNordic/Food-Recipe-MCP MCP server provides MCP clients with search access to approximately 50,000 recipes sourced from Food.com. The main use case is finding recipes from natural-language descriptions rather than relying only on exact keyword matches. Example searches include weeknight pasta, Swedish meatballs, high-protein chicken bowls, and Norwegian dishes such as kjøttkaker.

Search results can include the recipe title, description, preparation and cooking time, difficulty, diet information, main ingredient, servings, ingredients, instructions, nutrition, rating, rating count, source, and recipe ID. The underlying collection covers multiple cuisines, meal types, and cooking styles.

How it works

The hosted service processes each search through a two-stage retrieval pipeline. Dense vector search uses the intfloat/e5-large-v2 embedding model, while sparse retrieval uses BM25. Results from those methods are combined with Reciprocal Rank Fusion and then reordered by a cross-encoder reranker.

This approach supports semantic queries in both English and Norwegian. The search tool accepts an optional diet value, a maximum total cooking time in minutes, a difficulty value, and a result limit. Supported diet filters include vegetarian, vegan, gluten-free, dairy-free, low-carb, keto, and paleo. Difficulty can be set to easy, medium, or hard. The result limit defaults to five and can be increased to a maximum of 20.

Setup and configuration

AIDataNordic/Food-Recipe-MCP MCP server is available at https://recipes.aidatanorge.no/mcp using the streamable HTTP transport. Add that URL to an MCP client as a server named food-recipe, or register it with Claude Code using the HTTP transport. The README also provides a browser demo at https://recipes.aidatanorge.no/ for trying the service.

The hosted connection requires no local installation or configuration according to the project instructions. The repository describes the service as running with FastMCP over HTTP, Qdrant for its recipe collection, and a Cloudflare Tunnel for infrastructure access. Those implementation details are relevant if you are evaluating the project architecture, but the documented client connection uses only the remote MCP URL.

Tools and capabilities

The AIDataNordic/Food-Recipe-MCP MCP server exposes these tools:

• search_recipes: Searches the Food.com recipe collection with hybrid semantic and sparse retrieval, reranking, and optional diet, time, difficulty, and result-count filters.
• ping: Accepts a name and returns a response confirming that the recipe server is running.

Nutrition fields in returned recipes include calories, fat, protein, carbohydrates, sodium, fiber, and sugar per serving. Ratings include both the user rating and rating count when available.

Limitations and notes

The searchable data comes from Food.com and is described as approximately 50,000 recipes; it is not presented as a general-purpose recipe database beyond that source. Search results expose recipe information and metadata, but the provided material does not describe functions for creating, editing, saving, or rating recipes.

The documented remote endpoint is the supported connection method in the README. No API key, OAuth flow, or other credential configuration is specified. Although the project uses local components such as Qdrant and embedding models in its architecture, the material does not provide a verified command for running a local copy of AIDataNordic/Food-Recipe-MCP MCP server.