日本交通路线搜索MCP服务

navitime-mcp

A lightweight Model Context Protocol (MCP) server written in Rust that enables AI assistants to search transit routes across Japan — trains, subways, bullet trains, buses, walking, and more — powered by the Navitime API via RapidAPI.

What it does

Give any AI assistant (Claude, GPT, etc.) the ability to plan real transit trips in Japan:

User: "How do I get from Tokyo Station to Shibuya at 8am?"
AI:   [calls search_transit_route] →
      Route 1: JR Yamanote Line (local_train)
        Tokyo → Shibuya | 26 min | 200 yen (IC: 198 yen)

The server returns structured route data including fare breakdown, transfer count, walking distance, departure/arrival times, line colors, and company information.

Features

• Full Japan coverage — JR, private railways, Tokyo Metro, subways, Shinkansen, buses, domestic flights
• Rich route details — fares (cash + IC card), transfers, distances, travel times, platform info
• Transport metadata — line colors, company names, station IDs, node types
• MCP protocol — works with any MCP-compatible AI client over stdio (JSON-RPC 2.0)
• Hexagonal architecture — clean separation of domain, ports, adapters, and use cases
• Fully tested — unit tests with mocks + API response deserialization fixtures
• Fast — native Rust binary, async I/O with Tokio

Quick Start

1. Get a RapidAPI key

Subscribe to the Navitime Route Totalnavi API on RapidAPI (free tier available).

2. Build

git clone https://github.com/GHagui/mcp-navitime-rust.git
cd mcp-navitime-rust
cargo build --release

3. Configure your AI client

Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "navitime": {
      "command": "/path/to/target/release/navitime-mcp",
      "env": {
        "TOKEN_RAPIDAPI": "your-rapidapi-key"
      }
    }
  }
}

Claude Code

Add to .claude/settings.json or configure via Claude Code settings:

{
  "mcpServers": {
    "navitime": {
      "command": "/path/to/target/release/navitime-mcp",
      "env": {
        "TOKEN_RAPIDAPI": "your-rapidapi-key"
      }
    }
  }
}

Any MCP-compatible client

The server communicates over stdio using JSON-RPC 2.0. Set the TOKEN_RAPIDAPI environment variable and run the binary.

MCP Tools

search_transit_route

Finds train/walking transit routes in Japan between two geographic coordinates.

Example output

=== Route 1 ===
Total time: 13 min
Fare: 170 yen
IC Fare: 165 yen
Distance: 5900 m
Transfers: 0
Departure: 2025-10-01T08:01:00+09:00
Arrival: 2025-10-01T08:14:00+09:00

Sections:
  [Point] Omotesando
  [local_train] Tokyo Metro Ginza Line
    Line: Tokyo Metro Ginza Line
    Company: Tokyo Metro
    Depart: 2025-10-01T08:01:00+09:00
    Arrive: 2025-10-01T08:14:00+09:00
    Time: 13 min
  [Point] Ginza

Response data includes

• Summary: total time, distance, fare (normal + IC card), transfer count, walk distance
• Sections: alternating point/move segments with:
  • Station names, coordinates, node IDs
  • Transport details: line name, color, company, direction
  • Move type: walk, local_train, rapid_train, superexpress_train, domestic_flight, etc.
  • Timing: departure and arrival per segment

Architecture

src/
├── main.rs                          # Entry point & dependency injection
├── domain/
│   ├── models.rs                    # Route, Section, Transport, Fare types
│   └── error.rs                     # AppError enum (API, network, parsing, validation)
├── ports/
│   ├── inbound.rs                   # RouteSearchUseCase trait
│   └── outbound.rs                  # TransitApiClient trait (+ mock for tests)
├── use_cases/
│   └── search_route.rs             # Business logic + input validation + tests
└── adapters/
    ├── inbound/
    │   └── mcp_server.rs           # MCP JSON-RPC server (stdio) + response formatting
    └── outbound/
        └── rapidapi.rs             # Navitime RapidAPI HTTP client

The project follows Hexagonal Architecture (Ports & Adapters):

• Domain — Pure data models and error types, no external dependencies
• Ports — Trait interfaces defining inbound (use cases) and outbound (API clients) boundaries
• Use Cases — Application logic with coordinate validation; auto-defaults departure time to now
• Adapters — Concrete implementations: MCP server (inbound) and RapidAPI client (outbound)

Testing

cargo test

Tests include:

• Route search success flow with mock API client
• Input validation (invalid coordinates)
• API error propagation
• Default time generation when not provided
• JSON deserialization of real Navitime API response fixtures

Supported transport types

Requirements

• Rust 1.70+
• A RapidAPI key for the Navitime Route Totalnavi API

License

MIT