Kubishi Dictionary

An Owens Valley Paiute (Northern Paiute) language dictionary — a Vue 3 PWA deployed to Cloudflare Pages with D1 + Vectorize search.

Features

• 📖 Built from .lift XML files (SIL FieldWorks format)
• 🔍 Hybrid search: D1 FTS5 full-text + Vectorize semantic search when online
• 📴 Works offline with TF-IDF fallback from cached data (PWA)
• 🌓 Light/dark mode with persistent preference
• 📱 Fully responsive design
• 🎲 Random word discovery and word of the day
• 🤖 MCP server for AI assistant integration
• 🔗 REST API via Cloudflare Pages Functions

Setup

1. Install dependencies:

npm install

2. Place your .lift dictionary file in the project root as dictionary.lift

3. Build the data files and frontend:

npm run build

4. Seed the Cloudflare databases:

npx wrangler login                    # authenticate with Cloudflare
npm run seed:d1                       # seed D1 database
VECTORIZE_API_TOKEN=xxx node build-scripts/seed-vectorize.js  # seed Vectorize

5. Deploy:

npm run deploy

6. Preview locally:

npm run preview     # runs on :8788 with Cloudflare Functions

Updating Dictionary Data

When you receive a new .lift file:

npm run build:data       # 1. Parse .lift → JSON
npm run seed:d1          # 2. Re-seed D1 (drops and recreates tables)
VECTORIZE_API_TOKEN=xxx node build-scripts/seed-vectorize.js  # 3. Re-seed Vectorize
npm run deploy           # 4. Build and deploy

Project Structure

dictionary-static/
├── build-scripts/              # Build-time data processing & DB seeding
│   ├── parse-lift.js           # Parses .lift XML into JSON
│   ├── d1-schema.sql           # D1 database schema (FTS5, paiute_forms)
│   ├── seed-d1.js              # Seeds D1 from generated JSON
│   └── seed-vectorize.js       # Generates embeddings & seeds Vectorize
├── functions/api/              # Cloudflare Pages Functions
│   ├── _db.js                  # D1/Vectorize query helpers
│   ├── _shared.js              # Shared utilities (CORS, formatting)
│   ├── mcp.js                  # MCP server (JSON-RPC 2.0)
│   ├── search.js               # English search (FTS5 + Vectorize)
│   ├── search-paiute.js        # Paiute fuzzy search
│   ├── search-sentences.js     # Sentence search
│   ├── browse.js               # Alphabetical browse
│   ├── word/[id].js            # Word lookup
│   ├── word-of-the-day.js      # Word of the day
│   ├── random-word.js          # Random word
│   └── random-sentence.js      # Random sentence
├── src/
│   ├── views/                  # Page components
│   │   ├── Home.vue            # Search with Words/Sentences tabs
│   │   ├── Browse.vue          # Alphabetical browsing
│   │   ├── Word.vue            # Word detail page
│   │   ├── Pronunciation.vue   # Pronunciation guide
│   │   └── About.vue           # About page
│   ├── services/
│   │   ├── data.js             # Loads JSON data for offline use
│   │   ├── smart-search.js     # TF-IDF search (offline fallback)
│   │   ├── search-api.js       # Online search via API
│   │   └── utils.js            # Paiute fuzzy matching utilities
│   └── router/
│       └── index.js            # Vue Router configuration
├── public/data/                # Generated JSON (gitignored)
└── dictionary.lift             # Source dictionary file (place here)

Search Architecture

Online:  PWA/API/MCP → D1 FTS5 + Vectorize semantic search → hybrid scored results
Offline: PWA → local TF-IDF from cached words.json/sentences.json

• D1 FTS5: Full-text search with porter stemming on English definitions/glosses
• Vectorize: Semantic vector search using @cf/baai/bge-base-en-v1.5 embeddings (768 dims, cosine similarity)
• Hybrid scoring: score = (vectorize_similarity * 0.6) + (fts5_rank * 0.4)
• Paiute fuzzy: Phonetically-aware matching (k/g, t/d, s/z, u/u equivalence) with Levenshtein distance
• TF-IDF (offline): Term frequency-inverse document frequency with cosine similarity

Cloudflare Bindings

Binding	Service	Purpose
DB	D1 (kubishi-dictionary)	Word/sentence storage, FTS5 search
VECTORIZE_WORDS	Vectorize (kubishi-words)	Word semantic search (768 dims)
VECTORIZE_SENTENCES	Vectorize (kubishi-sentences)	Sentence semantic search (768 dims)
AI	Workers AI	Query embedding generation at search time

API

All endpoints are under /api/.

Endpoint	Description
GET /api/search?q=water	Search by English meaning/definition
GET /api/search-paiute?q=paa	Search by Paiute word (fuzzy matching)
GET /api/word/:id	Look up a word by ID
GET /api/browse?letter=T	Browse alphabetically by letter
GET /api/word-of-the-day	Get the word of the day
GET /api/random-word	Get a random word
GET /api/search-sentences?q=mountain	Search example sentences
GET /api/random-sentence	Get a random sentence

Common query parameters: limit (max results), skip (pagination offset).

MCP Server (AI Integration)

The dictionary exposes an MCP (Model Context Protocol) endpoint at /api/mcp, allowing Claude and other MCP-compatible AI assistants to search and browse the dictionary directly.

Connecting via Claude API

{
  "mcp_servers": [{
    "type": "url",
    "url": "https://kubishi.com/api/mcp",
    "name": "kubishi-dictionary"
  }],
  "tools": [{
    "type": "mcp_toolset",
    "mcp_server_name": "kubishi-dictionary"
  }]
}

Connecting via Claude Desktop

{
  "mcpServers": {
    "kubishi-dictionary": {
      "command": "npx",
      "args": ["mcp-remote", "https://kubishi.com/api/mcp"]
    }
  }
}

Available MCP Tools

Tool	Description
search_english	Search by English meaning or definition
search_paiute	Fuzzy search by Paiute word form
lookup_word	Get a word by its ID
browse_words	Browse entries alphabetically
get_letter_counts	Word counts per starting letter
word_of_the_day	Get today's featured word
random_word	Get a random dictionary entry
search_sentences	Search example sentences
random_sentence	Get a random example sentence

Testing with MCP Inspector

npx @modelcontextprotocol/inspector

Select Streamable HTTP and enter the MCP endpoint URL.

.lift File Format

This project expects a standard SIL FieldWorks .lift (Lexicon Interchange FormaT) XML dictionary file. The parser extracts:

• Lexical entries (words and headwords)
• Forms in multiple writing systems
• Senses with definitions and glosses
• Example sentences with translations
• Grammatical information (part of speech)
• Entry dates and GUIDs

Customization

Styling

Edit src/assets/style.css to customize colors and appearance. CSS variables make it easy:

• --bg-primary, --bg-secondary — Background colors
• --text-primary, --text-secondary — Text colors
• --accent-primary — Accent/link color
• All variables have separate dark mode values

Content

• Update src/views/About.vue with your credits and information
• Modify src/views/Pronunciation.vue for your language's pronunciation
• Edit acknowledgements in src/views/Home.vue

Credits

This site displays data from Glenn Nelson Jr.'s Owens Valley Paiute Dictionary, made possible through the contributions of native speakers and elders.

Built with Vue 3, Vite, Cloudflare D1/Vectorize/Workers AI, and custom hybrid search. Hosted on Cloudflare Pages.