MythxEngine

io.github.protoLabsAI由 GitHub 帳號 protoLabsAI 發布v0.3.0更新於 Oct 24, 2026

Tabletop RPG engine over MCP: dice, combat, stress, clocks, world generation, four bundled worlds

已驗證STDIO僅桌面

概覽

AI 產生的概覽

一款桌上角色扮演遊戲引擎,讓 AI 擔任遊戲主持人,提供確定性骰子、戰鬥、壓力、時鐘、世界生成與四個內建世界。

可以做什麼
  • 執行遊戲與角色:建立工作階段、載入世界概要、建立具生命值、壓力與裝備的角色
  • 判定與後果:依位置與效果擲骰判定、推動擲骰、抵抗後果、查詢規則
  • 戰鬥追蹤:開始戰鬥、擲先攻、攻擊,並記錄護甲、狀態與傷害
  • 時鐘與世界狀態:推進情境時鐘、追蹤調查與 NPC 關係、生成世界
  • 查詢並生成規則書內容,與引擎使用同一來源
適用情境
適合希望由助理主持或協助桌上角色扮演遊戲、並要求規則判定一致且可重現(同種子同結果)的情境,也適合體驗四個內建起始世界。
執行需求
以 stdio 在本機執行,可透過 npm 套件安裝(需 Node 22.13 或更新版本),或使用自帶 Node 執行環境的 Claude Desktop 擴充套件。選用環境變數 RPG_MCP_DATA_DIR 指定工作階段與世界包的存放位置;未宣告任何身分驗證。
安裝前請注意
工作階段與世界包會寫入本機資料目錄(預設位於使用者家目錄下),該目錄保存你的遊戲狀態。內建世界約 13 MB,首次執行時會匯入。README 說明此儲存庫是私有 monorepo 的下游鏡像。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 MythxEngine,將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

MythxEngine — tabletop RPG engine over MCP

[CI] [npm] [license: MIT] [node]

A complete tabletop RPG engine your AI can run: deterministic dice, five-tier outcomes, FitD-style stress, combat tracking, situation clocks, investigations, NPC relationships, world generation — 134 tools over the Model Context Protocol, with four complete starter worlds bundled in.

Claude (or any MCP client) becomes your Game Master. The engine keeps the rules honest: same seed + same actions = same outcome.

Install

Claude Code — plugin (skills + tools, recommended)

/plugin marketplace add protoLabsAI/mythx-mcp/plugin install rpg@mythx-plugins

One install delivers the MCP server plus play skills (/rpg:play, /rpg:combat-mode, /rpg:world-building-mode, …).

Claude Code — tools only

bash
claude mcp add rpg -- npx -y @mythxengine/mcp-server

Requires Node >= 22.13.

Claude Desktop

Download mythxengine-<version>.mcpb from the latest release and drag it into Settings → Extensions. No other setup — Claude Desktop ships its own Node runtime.

Five-minute walkthrough

Everything below is one conversation with Claude after installing. Four worlds are imported on first run — Port Miriam (harbor noir), Aethelgard (mythic frontier), Threshold Deep (weird depths), and the Treehouse Anthology.

1. Pick a world, make a character.

Start a session in Port Miriam. I want to play a dockside fixer who owes the wrong people money.

The GM calls create_session, load_world_summary, and create_character — your fixer gets HP, stress, gear, and a spot in the world's faction web.

2. Roll with real stakes. Every test is framed with position (how bad is failure?) and effect (how much does success buy?):

I try to slip past the harbor patrol to reach the smuggler's skiff.

roll_test → risky position / standard effect → 11 vs 13: partial"Yes, but..." — you reach the skiff, and a lantern swings your way.GM move suggested: put_someone_in_spot

Five outcomes — critical success, success, partial, failure, critical failure — and on a bad roll you can push_roll (spend 2 stress for +1d6) or resist_consequence to buy the severity down. Stress is a real currency; overflow means trauma.

3. Fight a combat round.

Two dockhands block the gangway. I draw my hook-knife.

start_combat → roll_initiative → attack — initiative order, armor, conditions, and damage all tracked; every attack returns the same five-tier outcome with suggested GM moves on a miss.

4. Watch a clock tick. The patrol you dodged? The GM started a harbor-alert clock. Fail forward enough times (tick_clock fires automatically on partials/failures when wired) and the dock lights come on — consequences arrive on schedule, not by fiat.

That's the loop: fiction in, mechanics out, consequences on clocks.

What's in the box

PieceWhat it gives you
packages/mcp-serverThe MCP server — 134 tools, published as @mythxengine/mcp-server
packages/enginePure, seeded game mechanics (dice, outcomes, combat)
packages/types / worlds / rulebookZod-validated schemas for every entity, plus rules lookup
packages/storageSQLite persistence (node:sqlite, zero native deps)
packages/tools / promptsTransport-agnostic tool implementations + runtime GM skills
world-builder/The rpg Claude Code plugin (skills, agents, hooks)
agent-kit/Portable GM system prompt + operating manual for any agent
docs/The rulebook + how the engine works
4 bundled worlds~13 MB of NPCs, factions, locations, situations, arcs — the demo

Rules & docs

  • docs/RULEBOOK.md — the complete rules, all 12 chapters (outcomes, position & effect, tests, stress, combat, conditions, gear, time & clocks, GM moves). Generated from packages/rulebook, the same source the lookup_rule and generate_rulebook tools serve — so it never drifts from the engine.
  • docs/engine.md — how the engine works: the five-tier outcome system, position & effect, stress, deterministic RNG, and how the packages fit together.
  • agent-kit/ — a portable GM system prompt and operating manual for running a game with any agent.

Want the full experience?

This repo is the engine. MythxEngine Desktop is the game: a visual frame-based play surface, scene imagery, generated world books, and polish. Follow along and get notified when it ships at mythxengine.com.

Development

bash
pnpm installpnpm build && pnpm test && pnpm lint && pnpm typecheck

See CONTRIBUTING.md — this repo is a downstream mirror of a private monorepo; issues are the fastest way to get fixes in.

License

MIT © protoLabs Studio

來源:README.md,提交 5c50567

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包,因此其工具未知。

版本歷史

1
  1. v0.3.0最新Sep 16, 2026

評分

已滿足適用檢查項的 50%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分