WayneCommand

Laya MCP — Development skill for Claude Code

Development community

How to install Laya MCP

This entry records only its repository, not the path inside it, so there is no exact command to give. Open WayneCommand/laya-mcp and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Laya MCP does

A single forward pass answers a set of "typed questions," returning calibrated probabilities alongside a decision contract ready for direct branching execution—enabling any MCP-compatible agent (such as Claude Code, Codex, Cursor, or custom-built agents) to make automated decisions within milliseconds.

Alternatives in Development

Notebooklm Skill — Use this skill to enable Claude Code to communicate directly with your Google NotebookLM notebooks 5.9k ★
Om Meeting — Prep for any meeting by topic — subject-forward briefing with open tasks, blockers, and brainstormed considera 4.6k ★
Kimi CLI Agent — Answers one council question as plain text, with no tool access 664 ★

README

laya-mcp

**把 Laya System-1 决策模型封装成一个工程化的 MCP Server（FastAPI 宿主）**

一次前向传播回答一组「带类型的问题」，返回**校准概率** + **可直接分支执行的决策契约**，让任意支持 MCP 的 Agent（Claude Code / Codex / Cursor / 自研 Agent）都能在**毫秒级**完成自动决策。

[![Python](https://img.shields.io/badge/python-3.13-3776AB?logo=python&logoColor=white)](https://www.python.org/) [![FastAPI](https://img.shields.io/badge/FastAPI-0.141-009688?logo=fastapi&logoColor=white)](https://fastapi.tiangolo.com/) [![MCP](https://img.shields.io/badge/MCP-2.2-000000)](https://modelcontextprotocol.io/) [![Laya](https://img.shields.io/badge/laya-0.3.5-FF6F00)](https://huggingface.co/convaiinnovations/laya)

这是什么

`laya-mcp` 把 [Laya](https://huggingface.co/convaiinnovations/laya)（一个 **System-1** 决策模型族）包装成一个**常驻服务**，同时对外暴露两种接口：

接口	地址	面向谁
MCP Streamable HTTP	`http://127.0.0.1:8077/mcp`	任意 MCP 客户端 / Agent
REST API	`http://127.0.0.1:8077/v1/...`	脚本、看板、健康检查、非 MCP 调用方
OpenAPI 文档	`http://127.0.0.1:8077/docs`	人

它不是「又一个 LLM 接口」。Laya 是**非生成式**的：一次 forward pass 回答**一组结构化问题**，输出的是**校准过的概率分布**，而不是一段文本。这使得它可以在 10~50ms 内做出「该走哪条路」的判断，成本约为调用一次完整 LLM 的千分之一。

与传统做法对比

`laya-test.py`（见 [`examples/quickstart_router.py`](examples/quickstart_router.py)）演示了最小用法：加载 Router → 构造问题 → `router.predict()` → 用阈值手写 `if`。

这个工程把它**产品化**了：

关注点	`laya-test.py`	`laya-mcp`
模型驻留	每次进程启动重新加载	常驻单例 + 启动预热（warmup）
阈值策略	每个调用方自己写 `if conf >= 0.8`	决策契约，由服务统一输出 `grade`
并发	单线程阻塞	线程池 + 事件循环不阻塞