pm-minecraft
Enables MCP clients to control a Minecraft survival character via Mineflayer, with a web UI, viewer, and custom TypeScript script execution.
README
pm-minecraft
A self-contained Minecraft survival body for MCP clients.
It runs Mineflayer, Prismarine Viewer, a local web UI, and a Streamable HTTP MCP server.
It does not contain an agent, model, or cognitive runtime. Only some minimal instruction telling your agent of choice how to use the MCP, look at screenshots and make custom TypeScript scripts, that can be executed via MCP.
Huge thanks to https://github.com/minedojo/voyager and https://github.com/Mega-Gorilla/Discovery :3
This is part of an ongoing effort on my side to make a fun cognitive architecture AI companion that can play minecraft with you. Also works as standalone ^_^
Setup
Requirements: Windows PowerShell, Node.js 20+, Python 3.12 through py, and a
reachable Minecraft Java 1.19.x server with the character in survival mode.
Set-Location C:\workspace\pm-minecraft-mcp
.\setup.ps1
The setup follows the shared PM workflow: it uses uv to create this
repository's Python 3.12 .venv, syncs its lockfile, and installs locked Node
packages. scripts/setup.ps1 remains as a compatibility wrapper.
If it doesn't work tell your coding agent to fix it.
Minecraft
I stole the setup from https://github.com/Mega-Gorilla/Discovery.
I have never modded Minecraft manually, what works for me:
- Download Prism
- Install 1.19.4
- Install Fabric Loader 0.19.3
- Install these mods via Prism:
- Fabric API
- CompleteConfig
- Mod Menu
- Multiplayer Server Pause (Forge)
- Make Survival World, Cheats Enabled, Peaceful
- Enter and "Open to LAN" on Port 12345
Create and start a character
.\scripts\init_character.ps1 `
-Name Floppa `
-AgentRoot C:/Temp/Floppa `
-ArtifactRoot C:/Temp/Floppa/artifacts/minecraft
.\scripts\start_minecraft_mcp.ps1 `
-Name Floppa `
-MinecraftHost 127.0.0.1 `
-MinecraftPort 12345 `
-AgentRoot C:/Temp/Floppa `
-ArtifactRoot C:/Temp/Floppa/artifacts/minecraft
The initializer creates the agent workspace, memory/minecraft/, drafts/,
skills/, lib/minecraft.ts, and .mcp.json. The launcher prints the local
web UI, Prismarine viewer, and MCP URLs. For multiple characters, use unique
-WebPort, -ViewerPort, and -McpPort values.
Stop an instance with:
.\scripts\stop_minecraft_mcp.ps1 -ArtifactRoot C:/Temp/Floppa/artifacts/minecraft
Then use it from the workspace:
Set-Location C:/Temp/Floppa
codex
Codex reads .mcp.json. Prompt it to use the minecraft server; state logs
and screenshots are under ./artifacts/minecraft.
Perception and viewer modes
minecraft_find_block defaults to require_visible: true. This is the normal
survival setting: it returns only blocks with an unobstructed ray from the
character's head. An agent must explore, move to a better viewpoint, or use a
different observation when it cannot see the target. For long-range planning,
set require_visible: false; the result is a loaded-world location only and
must still be reached and visibly verified before mining.
Cool stuff
Here is the WebUI where you can manually control the character or look at how the coding agent uses the MCP.
And here is Codex describing my character's skin. Prismarine only renders default Steve :(
Development
npm test
npm run build
.\.venv\Scripts\python.exe -m compileall mcp
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。