MCP Michelin Guide Server

About

An MCP (Model Context Protocol) server that lets you search Michelin Guide restaurants worldwide from Claude Code.

Powered by a local SQLite database with the FTS5 full-text search engine for fast, accurate queries -- with a web fallback to supplement the latest information from the official Michelin Guide site.

Key Features

• Multi-criteria restaurant search -- Combine free-text search, city, country, cuisine type, award level, Green Star, and price level
• Restaurant detail lookup -- Get 17-field details by name (partial match) or Michelin Guide URL
• Web fallback search -- Fetches the latest info from the Michelin Guide website when local data is insufficient
• ~18,900 restaurants worldwide -- Based on open-source data updated monthly
• Local-first architecture -- Fast search with SQLite + FTS5, no network required

Architecture

flowchart LR
    subgraph "Data Pipeline"
        CSV["CSV Source Data<br/>ngshiheng/michelin-my-maps"]
        SETUP["src/scripts/setup-data.ts"]
        DB["data/michelin.db<br/>SQLite + FTS5"]
        CSV -->|Download & Parse| SETUP -->|Create Tables & Index| DB
    end

    subgraph "MCP Server"
        INDEX["src/index.ts<br/>stdio transport"]
        DB_MOD["src/db.ts<br/>Query Engine"]

        subgraph "Tools"
            T1["search_restaurants<br/>Multi-criteria Search"]
            T2["get_restaurant_detail<br/>Detail Lookup"]
            T3["search_latest_info<br/>Web Fallback"]
        end

        INDEX --- T1 & T2 & T3
        T1 & T2 -->|Query| DB_MOD
        T3 -->|Query| DB_MOD
    end

    DB_MOD -->|Local Query| DB
    T3 -->|Fallback| WEB["guide.michelin.com"]

    CLAUDE["Claude Code"] <-->|MCP stdio| INDEX

Prerequisites

Item	Version	Note
Node.js	>= 18	LTS recommended
npm	>= 9	Included with Node.js
Claude Code	Latest	MCP client

Installation

1. Clone the repository

git clone https://github.com/anhyobin/michelin-guide-mcp.git
cd michelin-guide-mcp

2. Install dependencies

npm install

3. Initialize the database

Downloads the CSV source data and creates the SQLite database.

npm run setup

This command will:

1. Download the latest CSV from the ngshiheng/michelin-my-maps repository
2. Parse the CSV and clean the data
3. Create data/michelin.db with the restaurants table + FTS5 virtual table

4. Build

npm run build

Claude Code Integration

Option 1: Using .mcp.json (Recommended)

Create or edit a .mcp.json file in your project root:

{
  "mcpServers": {
    "michelin-guide": {
      "command": "node",
      "args": ["build/index.js"],
      "cwd": "/absolute/path/to/michelin-guide-mcp"
    }
  }
}

Replace cwd with the absolute path to your cloned project directory.

Option 2: Register via Claude Code CLI

claude mcp add michelin-guide -- node /absolute/path/to/michelin-guide-mcp/build/index.js

Verify the integration

Ask Claude Code a question to confirm the MCP server is working:

Find 3-star Michelin restaurants in Tokyo

MCP Tool API Reference

1. search_restaurants -- Multi-criteria restaurant search

Search restaurants by combining multiple criteria. Results are sorted by award level, returning 20 per page.

Parameter	Type	Required	Description
query	string	-	Free text search (FTS5 full-text search)
city	string	-	City name
country	string	-	Country name
cuisine	string	-	Cuisine type (e.g., Japanese, French, Italian)
award	string	-	Award level: 1 Star | 2 Stars | 3 Stars | Bib Gourmand | Selected Restaurants
green_star	boolean	-	Whether the restaurant has a Michelin Green Star
price_level	number	-	Price level (1=budget to 4=luxury)
page	number	-	Page number (default: 1)

Example response:

{
  "total": 42,
  "page": 1,
  "totalPages": 3,
  "results": [
    {
      "name": "Sukiyabashi Jiro",
      "award": "3 Stars",
      "city": "Tokyo",
      "country": "Japan",
      "cuisine": "Sushi",
      "price_level": 4,
      "green_star": false,
      "url": "https://guide.michelin.com/..."
    }
  ]
}

2. get_restaurant_detail -- Restaurant detail lookup

Look up full details (17 fields) for a restaurant by name or Michelin Guide URL. Returns a candidate list when multiple matches are found.

Parameter	Type	Required	Description
name	string	Conditional	Restaurant name (partial match supported)
url	string	Conditional	Michelin Guide URL (exact match)

Either name or url must be provided.

Response fields:

Field	Description
name	Restaurant name
address	Address
location	Location info (city, country)
longitude	Longitude
latitude	Latitude
city	City
country	Country
price	Price info (original)
price_level	Price level (1--4)
cuisine	Cuisine type
award	Award level
green_star	Green Star status
url	Michelin Guide URL
description	Description
phone_number	Phone number
website_url	Website
facilities_and_services	Facilities and services

3. search_latest_info -- Web fallback search

Fetches the latest information from the Michelin Guide website when local database data is insufficient.

Parameter	Type	Required	Description
restaurant_name	string	Yes	Restaurant name
location	string	-	Location hint (city, country, etc.)

Usage Examples

Ask Claude Code in natural language and the MCP tools are called automatically.

Restaurant search

Find Michelin-starred restaurants in Seoul

-> search_restaurants: { "city": "Seoul" }

Which 3-star French restaurants in France also have a Green Star?

-> search_restaurants: { "country": "France", "cuisine": "French", "award": "3 Stars", "green_star": true }

Find affordable Bib Gourmand Italian restaurants

-> search_restaurants: { "cuisine": "Italian", "award": "Bib Gourmand" }

Detail lookup

Tell me about Sukiyabashi Jiro

-> get_restaurant_detail: { "name": "Sukiyabashi Jiro" }

Show me this restaurant's info: https://guide.michelin.com/...

-> get_restaurant_detail: { "url": "https://guide.michelin.com/..." }

Latest info

Get the latest info for a restaurant in Ginza, Tokyo

-> search_latest_info: { "restaurant_name": "...", "location": "Tokyo Ginza" }

Updating Data

The source data is updated monthly from ngshiheng/michelin-my-maps. To refresh:

# Remove existing database and recreate
rm -rf data/
npm run setup

Project Structure

michelin-guide-mcp/
├── .mcp.json                     # Claude Code MCP server config
├── package.json                  # Project metadata and dependencies
├── tsconfig.json                 # TypeScript configuration
├── .gitignore
├── data/
│   └── michelin.db               # SQLite DB (gitignored, generated)
├── src/
│   ├── index.ts                  # MCP server entry point (stdio)
│   ├── db.ts                     # DB initialization and queries
│   ├── types.ts                  # Type definitions
│   ├── utils.ts                  # Utility functions
│   ├── scripts/
│   │   └── setup-data.ts         # CSV download -> SQLite DB setup
│   └── tools/
│       ├── search-restaurants.ts # Multi-criteria search tool
│       ├── get-restaurant-detail.ts # Detail lookup tool
│       └── search-web.ts         # Web fallback search tool
└── build/                        # Compiled JavaScript output

Tech Stack

Technology	Purpose
TypeScript 5.8	Type-safe server development
@modelcontextprotocol/sdk	MCP server framework
better-sqlite3	High-performance native SQLite bindings
SQLite FTS5	Full-text search engine
csv-parse	CSV data parsing
Zod	Runtime schema validation

License

This project is distributed under the MIT License.

Credits

• Restaurant data: ngshiheng/michelin-my-maps (MIT License)
  • Open-source project providing data on ~18,900 Michelin Guide restaurants worldwide
  • Automatically updated monthly for the latest data
• MCP Protocol: Model Context Protocol by Anthropic