mc-skill-library
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mc-skill-librarysearch the skill library for building a shelter and run it"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mc-skill-library
为 LLM Agent 设计的技能库:语义检索 + 可执行代码 —— 让 Agent 学会的技能终身复用,零 LLM 调用。
这是什么
LLM Agent 玩 Minecraft 时,每个动作都过一遍 LLM,延迟 1-10 秒,体验割裂。 本项目的思路(源自 Voyager):把 Agent 学到的能力 沉淀为可执行代码(技能),向量化入库;新任务来了语义检索复用,命中即零 LLM 调用。
"帮我弄个住的地方" ──► skill_search ──► build_shelter.js ──► skill_exec
(语义检索,~1ms) (技能库命中) (流畅执行,无 LLM)Related MCP server: Minecraft MCP Server
架构
任意 MCP 客户端 mc-skill-library (本仓库)
(x-code / Claude Desktop / Cursor)
│ MCP stdio ┌────────────────────────────┐
├─ skill_save ─────────────►│ 入库: 描述 → embedding │
├─ skill_search ───────────►│ 检索: 语义 top-k + 元数据过滤│
├─ skill_exec ─────────────►│ 执行: node 子进程 + 看门狗 │
└─ skill_stats ────────────►│ 统计: 命中/成功率(排序加权) │
│ │
│ 检索引擎: numpy 暴力精确检索 │
│ (≤1万级技能 0.1ms,够用且更对)│
└────────────────────────────┘为什么不用向量数据库/HNSW:技能库量级 ≤1k,1024 维暴力检索 ~0.1ms,
占端到端延迟(embedding 300ms + LLM 秒级)的万分之一。
生产路径 = 最简单、精确、过滤友好的方案;ANN 从零实现作为独立实验项目
测「从什么规模开始才值得用」(见 bench/)。
快速开始
git clone https://github.com/Steve-pixel-cpu/mc-skill-library
cd mc-skill-library
uv sync # 或 pip install -e .
# 直接跑(开发模式)
python server.py配置到任意 MCP 客户端(以 Claude Code 风格配置为例):
{
"mcpServers": {
"skills": {
"command": "python",
"args": ["path/to/mc-skill-library/server.py"]
}
}
}MCP 工具
工具 | 说明 |
| 保存技能:名称+描述+JS 代码,自动 embedding 入库 |
| 语义检索:任务描述 → top-k 技能(支持元数据过滤) |
| 执行技能:node 子进程跑 JS,带超时与看门狗 |
| 库统计:技能数/命中/成功率 |
路线图
阶段 0: MCP server 骨架 + JSONL 存储 + 关键词检索(能跑)
阶段 1: 接入 Minecraft bot,种子技能 5 个
阶段 2: embedding 语义检索 + 混合路由(结构化查字典/语义走向量)
阶段 3: benchmark(暴力 vs chromadb + 规模扫描,交叉点分析)
选修: 自研 HNSW 补入对比曲线
可选: 事件桥(MC 事件唤醒 Agent)+ 技能自动沉淀
已定边界
只支持官方 Java 版(锁 1.21.x)+ 离线模式;不进第三方反作弊服务器
不做战斗/实时反应(那是 Mineflayer 行为树的课题,不是本项目的)
技能载体是 mineflayer JS;检索/存储/执行器全 Python
License
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
Private, portable memory and reusable skills for AI agents.
Memory system for AI agents with semantic search. Store and recall memories with ease.
Capability registry for the agentic economy. Semantic search over verified MCP server listings.
Persistent memory and knowledge management for AI agents with semantic search and 50+ tools.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to play and interact with Minecraft servers through mineflayer, providing automated actions like mining, movement, crafting, and real-time game event monitoring.7 npm8MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to control Minecraft bots through over 30 verified skills including navigation, resource gathering, and advanced building. It allows for multi-bot management and seamless integration with MCP clients to automate complex in-game tasks.76MIT
- AlicenseAqualityAmaintenanceCognitive memory for AI agents. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client.21122 PyPI24MIT
- AlicenseNot gradedqualityCmaintenanceAn MCP server that gives AI agents a shared, living memory by distilling solved tasks into reusable, verifiable skills, enabling collective intelligence and reducing redundant inference.0MIT