Reverse-engineered API reference for Hytale modding, derived from the server JAR. Served as a static site and queryable via MCP.

Live site: hydex.kazyyk.dev

Live MCP server: mcp.hydex.kazyyk.dev/mcp (/sse also available)

What this is: A reverse-engineered API reference derived from the Hytale server JAR. A four-phase tooling pipeline decompiles, classifies, and maps the server's internal APIs into a structured, searchable reference for mod developers.

What this is not: Official documentation, manually authored content, or a replacement for Hypixel Studios' own documentation efforts. Every documented type, method, and field is mechanically derived from the game files.

HytaleServer.jar
 │
 ▼
┌─────────────────────────────────────────────┐
│ Pipeline (Phases 1-4) │
│ Decompile → Classify → Map → Generate │
└─────────────────┬───────────────────────────┘
 │
 ▼
 output/docs/
 65 markdown files
 ┌───────┴───────┐
 ▼ ▼
 Starlight Site R2 Bucket
 (Cloudflare (hytale-modding-corpus)
 Pages) │
 ▼
 AI Search
 (Vectorize + Workers AI)
 │
 ▼
 MCP Worker
 (Cloudflare Workers)
 │
 ▼
 MCP Client

See spec/generator-spec.md for the full engineering specification.

65 markdown files organized by system:

• 31 Events — Player, block, entity, chat, command, and server lifecycle events with dispatch context, parameters, and cancellation behavior
• 22 Key Classes — Plugin entry points, command framework, registries, ECS components, and server internals
• 3 JSON Schemas — Block, item, and crafting recipe data asset schemas with field types, defaults, and validation rules
• Section indexes — Overview pages for each category including command system reference with 55+ built-in commands

Every file includes structured frontmatter (kind, FQCN, package, API surface, related types, tags) that drives site navigation and RAG chunk metadata.

Built with Astro Starlight and deployed to Cloudflare Pages. The site syncs from output/docs/ at build time via a script that copies files and patches the landing page with Starlight's splash template.

Deployed automatically on push via Cloudflare Pages git integration.

A remote MCP server on Cloudflare Workers exposes the documentation corpus via semantic search. It uses Cloudflare AI Search (Vectorize + Workers AI) for retrieval with query rewriting and BGE reranking, returning raw document chunks for the client to process.

The MCP endpoint is https://mcp.hydex.kazyyk.dev/mcp (HTTP transport). An SSE endpoint is also available at /sse.

Most MCP-compatible clients can connect using the HTTP URL directly. Examples:

Claude Code:

claude mcp add --transport http hydex https://mcp.hydex.kazyyk.dev/mcp

Claude Desktop (claude_desktop_config.json):

{
 "mcpServers": {
 "hydex": {
 "type": "http",
 "url": "https://mcp.hydex.kazyyk.dev/mcp"
 }
 }
}

See mcp/README.md for setup, deployment, and design decisions.

A GitHub Action (.github/workflows/sync-docs-r2.yml) syncs output/docs/ to the R2 bucket on push to main, excluding JSON metadata files. AI Search re-indexes automatically when the bucket changes.

hydex/
├── AGENTS.md # Agent instructions (CLAUDE.md symlinks here)
├── CONTRIBUTING.md # Contribution guidelines
├── LICENSE # Project license
├── README.md # This file
├── SECURITY.md # Security policy
├── spec/
│ └── generator-spec.md # Engineering specification
├── input/ # Place HytaleServer.jar here
├── artifacts/ # Pipeline intermediate outputs
│ ├── decompiled/ # (gitignored)
│ ├── class-index.json # Phase 1: structured class index
│ ├── surface.json # Phase 2: API surface classification
│ ├── internal-index.json # Phase 2: internal types
│ ├── surface-review.json # Phase 2: borderline cases
│ ├── systems.json # Phase 3: mapped systems
│ └── cross-refs.json # Phase 3: cross-references
├── output/
│ └── docs/ # Phase 4: generated documentation (65 files)
├── tools/ # Phase 1-2 CLI tools (Java + Gradle)
│ ├── run.sh # Phase 1 entry point
│ ├── classify.sh # Phase 2 entry point
│ ├── gradlew # Gradle wrapper
│ └── app/ # Java source
├── site/ # Documentation site (Astro Starlight)
│ ├── astro.config.mjs
│ ├── scripts/sync-docs.mjs
│ ├── src/
│ │ ├── content.config.ts # Extended frontmatter schema
│ │ └── styles/custom.css
│ └── package.json
├── mcp/ # MCP server (Cloudflare Workers)
│ ├── src/index.ts # McpAgent with search_docs tool
│ ├── wrangler.jsonc
│ └── package.json
└── .github/
 └── workflows/
 └── sync-docs-r2.yml # R2 bucket sync on push

The pipeline has already been run and the outputs are committed. To re-run from a new JAR:

# Phase 1: Decompile & index
cd tools && ./run.sh ../input/HytaleServer.jar
# Phase 2: Classify API surface
cd tools && ./classify.sh
# Phases 3-4: See AGENTS.md for instructions
# Build site locally
cd site && npm install && npm run dev
# Deploy MCP server
cd mcp && npm install && npx wrangler deploy

Built with Vineflower, JavaParser, Astro Starlight, Cloudflare Workers, and Claude Code.

This project generates documentation from Hytale's game files for community reference. Hytale is a trademark of Hypixel Studios. This project is not affiliated with or endorsed by Hypixel Studios.