Food-Recipe-MCP is a production-grade food recipe semantic search server for AI agents and large language models (LLMs), built on the Model Context Protocol (MCP).
Main capabilities:
- Performs semantic search over more than 50,000 recipes from Food.com.
- Uses two-stage hybrid retrieval: dense vector retrieval and sparse BM25 retrieval, fused via Reciprocal Rank Fusion (RRF), then reranked through a cross-encoder.
- Queries support natural language in English and Norwegian.
- Can filter by diet, cooking time, and difficulty.
- Recipe data includes per-serving nutrition information (calories, fat, protein, carbohydrates, sodium, fiber, sugar) as well as user ratings and rating counts.
- Recipes cover a wide range of cuisines, meal types, and cooking styles.
How to use:
- Can be integrated through MCP client configuration (the official text mentions "Add to your MCP client config" and "Or with Claude Code", but does not provide the specific configuration content or installation commands, so they are not provided here).
- Provides an online endpoint:https://recipes.aidatanorge.no/mcp，传输方式为 streamable-http.
- Provides an online demo:https://recipes.aidatanorge.no/ , which can be tried in a browser without installation or configuration.
Example queries:
- "Swedish meatballs with gravy"
- "healthy high-protein chicken bowl"
- "easy chocolate cake for beginners"
- "traditional Norwegian kjøttkaker"
- "hurtig pasta med kylling"
Tech stack:
- Embeddings: intfloat/e5-large-v2 (1024-dimensional dense) + Qdrant/bm25 (sparse)
- Reranker: cross-encoder/mmarco-mMiniLMv2-L12-H384-v1
- Vector database: Qdrant (self-hosted)
- Server: FastMCP 3.2 over HTTP
- Infrastructure: Ubuntu Server 24 LTS, Cloudflare Tunnel
Use cases:
- Developers who need to integrate recipe retrieval capabilities into AI agents or LLM applications.
- Scenarios that require filtering recipes by diet, cooking time, and difficulty and obtaining nutrition information.
- Recipe search scenarios that require support for natural language queries in English or Norwegian.
Limitations:
- The official text does not state pricing, fees, platform, performance metrics, or specific usage step details, so this information is not provided here.
- The data source is Food.com, with about 50,000 recipes.
