lua-ai-global

lua-platform-mcp

Community lua-ai-global
Updated

Read-only MCP server for the Lua AI platform (heylua.ai) — list agents, read primitive versions, view deployment status, tail logs. Works with Cline, Cursor, Claude Code, Codex CLI, and any MCP-aware client.

lua-platform-mcp

Read-only MCP server for the Lua AI platform — exposes 5 tools that let any MCP-aware client (Cline, Cursor, Claude Code, Codex CLI, Continue, Zed, etc.) query Lua agent state mid-conversation.

Tools exposed

Tool What it returns
list_agents All agents accessible to the authenticated user ({id, name, orgId, orgName}[])
get_agent Public-facing config for one agent by ID
list_primitive_versions Versions of a skill / webhook / job / preprocessor / postprocessor / persona
get_deployment_status What's live in production right now (composed from per-type version endpoints)
tail_logs Structured log fetch (mirrors lua logs --json)

All tools are read-only — this server cannot push, deploy, or modify any state.

Install

Cline / any MCP client (recommended)

After cloning + building, register the server in your client's MCP config. For Cline, that's the cline_mcp_settings.json:

{
  "mcpServers": {
    "lua-platform": {
      "command": "node",
      "args": ["/absolute/path/to/lua-platform-mcp/dist/server.js"],
      "env": {
        "LUA_API_KEY": "lk_..."
      }
    }
  }
}

From source

git clone https://github.com/lua-ai-global/lua-platform-mcp
cd lua-platform-mcp
npm install
npm run build
# Now dist/server.js is the entry point

From npm (once published)

npx lua-platform-mcp   # one-shot
# or, in MCP config:
# "command": "npx", "args": ["-y", "lua-platform-mcp"]

Authentication

The server reads your Lua API key in this order:

  1. LUA_API_KEY environment variable
  2. ~/.lua-cli/credentials (written by lua auth configure)
  3. .env file in CWD

Get a key from admin.heylua.ai (sign up + create an agent → Settings → API Keys), or use the Lua CLI:

npm install -g lua-cli
lua auth configure

Verify it works

After registering, restart your MCP client and ask: "What lua-platform tools do you have?"

You should see all 5 tools listed under the mcp__lua-platform__* prefix (Cline / Cursor / Claude Code) or as direct names (Codex / Continue).

Try: "List my Lua agents." → should return your agents, IDs, and orgs.

Architecture

  • src/server.ts — MCP stdio bootstrap (uses @modelcontextprotocol/sdk)
  • src/tools/*.mjs — one file per tool, each exporting { name, description, inputSchema, handler }
  • src/auth.mjs — credential resolution (mirrors lua-cli's chain)
  • src/api-client.mjs — typed HTTP wrapper around https://api.heylua.ai
  • dist/server.js — esbuild-bundled single-file output (built via npm run build)

Why "read-only"?

Mutating operations (lua push, lua deploy, etc.) live in the Lua CLI directly, behind the §3.3 deploy-safety contract that requires explicit user confirmation. We deliberately don't expose them via MCP because MCP tool calls don't carry the human-confirmation step. Read tools are safe; write tools belong in the CLI.

If you want a richer "build agents inside your IDE" experience (slash commands, scaffolding, deploy gates), install one of the IDE-specific Lua plugins instead — they bundle this MCP server alongside the workflow:

License

MIT © Lua AI

MCP Server · Populars

MCP Server · New

    hsingjui

    ContextWeaver

    ContextWeaver 是一个基于 MCP 协议、利用 Tree-sitter 和向量搜索为大语言模型提供本地代码库智能上下文编织与检索的工具。

    Community hsingjui
    qase-tms

    Qase MCP Server

    An official Qase MCP server

    Community qase-tms
    repowise-dev

    repowise

    Codebase intelligence for AI-assisted engineering teams: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

    Community repowise-dev
    wwwzhouhui

    即梦 MCP 服务器

    一个为即梦AI打造的MCP服务器,让Claude、Cherry Studio等AI应用直接调用即梦的AI生成能力。基于jimeng-free-api-all开源项目,提供OpenAI兼容接口。 核心功能:文本生成图像(即梦4.0/3.1)、图像合成(多图融合)、文本生成视频(480p-1080p)、图像生成视频(静态转动态)。 支持三种模式:stdio(Claude Desktop)、SSE(Web)、HTTP REST API(跨平台)。Docker一键部署,开箱即用。异步轮询优化,确保长时间任务稳定完成。 需要Python 3.10+和Docker,配置SessionID即可使用,每日免费66积分。适合AI创作者、开发者学习MCP协议。MIT开源,代码透明。

    Community wwwzhouhui
    kaorii-ako

    栞 Shiori

    Open-source AI study companion — Google Classroom sync, Gemini AI plans, SRS flashcards, GPA predictor, AI quiz, MCP server for Claude Code. Try demo at shiori-v1.vercel.app

    Community kaorii-ako