1. Conduid
  2. Communication
  3. AI Mind
MCP server · Communication

AI Mind

一个按版本持续演进的 AI Native Runtime Skeleton。 它从本地聊天闭环起步,逐步演进到: 结构化流式协议 Tool Calling Multi-Tool Runtime Skill Runtime MCP / Agent / 数据层

Unclaimed llmmcpskillstools
34Low

Scored 4 months ago · breakdown

About AI Mind

AI Mind is an MCP server in the Communication category: 一个按版本持续演进的 AI Native Runtime Skeleton。 它从本地聊天闭环起步,逐步演进到: 结构化流式协议 Tool Calling Multi-Tool Runtime Skill Runtime MCP / Agent / 数据层. It has been installed 0 times through Conduid.

Install

Clone
git clone https://github.com/HWYD/ai-mind

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

Ask AI about AI Mind

Powered by Claude · Grounded in docs

I know everything about AI Mind. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ·README presentNot checked yet.
  • ·License declaredNot checked yet.
  • ·Tests presentNot checked yet.
  • ·Dependencies pinnedNot checked yet.
  • ·No dynamic code executionNot checked yet.
  • ·Scoped permissionsNot checked yet.

README

AI Mind

一个按版本持续演进的 AI Native Runtime Skeleton。

它不是一次性做完的产品,而是一套逐步生长的运行时骨架:

  • 本地聊天闭环
  • 结构化流式协议
  • Tool Calling
  • Multi-Tool Runtime
  • Skill Runtime
  • MCP 能力接入
  • Agent / 数据层等更长线能力

系列博客

当前状态

当前版本:v0.0.11

这版的主线不是继续重构聊天主链,而是在 v0.0.10 已稳定的 facade / runtime / stream-core 基础上,补齐 capability model、skill metadata、local/remote MCP 能力面,并验证单个 remote MCP server 的最小闭环:

  • 建立统一 Capability Model,覆盖 Tool / Resource / Prompt
  • 扩展 reader-skill / utility-skill 的结构化 skill metadata
  • 通过本地 MCP 暴露 local-file-summary Prompt
  • 新增独立服务 project-assistant-service,验证 remote MCP Resource / Prompt / Tool
  • 补齐最小 capability runtime 消费闭环
  • 前端展示 Skill 命中、capability 类型、local / remote 来源和 server 信息

当前能力

Chat Runtime

  • LangChain.js + Ollama
  • NDJSON 流式协议
  • reasoning / tool / resource / text 多段式消息流
  • skill / prompt 执行事实展示
  • 统一 error chunk 语义
  • authoritative answer 运行时策略
  • 最近 N=8 轮上下文
  • 最小 capability runtime 消费闭环

Skills

  • utility-skill
  • reader-skill

Tools

  • calculator
  • datetime
  • text-transform
  • unit-convert
  • city-weather
  • local-text-read

MCP Host MVP

  • @modelcontextprotocol/sdk
  • 本地 stdio MCP Host
  • remote Streamable HTTP MCP Host
  • weather-server
  • project-files-server
  • project-assistant-service
  • MCP Tool / MCP Resource adapter
  • MCP Prompt adapter

工程化边界

  • route -> chat-service facade -> runtime -> skills / tools / mcp
  • @ai-mind/stream-core / packages/stream-core 负责稳定流式内核
  • apps/webapp/tests/** 为唯一 webapp 自动化测试目录
  • packages/stream-core/tests/** 为 package 测试目录

当前结构

Webapp

  • apps/webapp/app/api/chat/route.ts
    • HTTP 边界与错误映射
  • apps/webapp/lib/ai/chat-service.ts
    • 薄 facade,负责创建内部流、构造中间 StreamResult 并包装 Response
  • apps/webapp/lib/ai/runtime/
    • 正式聊天运行时编排层
  • apps/project-assistant-service/
    • NestJS remote MCP 服务,当前用于验证单 server 最小闭环
  • apps/webapp/tests/
    • Webapp 自动化测试

Stream Core Package

  • packages/stream-core/src/protocol/
    • ChatStreamChunk 与错误协议类型
  • packages/stream-core/src/core/
    • lifecycle、error helper、static part writer
  • packages/stream-core/src/adapters/web/
    • NDJSON writer
  • packages/stream-core/tests/
    • package 单测

v0.0.11 的关键判断

这版有三个重要原则:

  1. capability model 是统一描述层,不是统一执行链
  2. Skill metadata 只描述承接范围和输出风格,不扩张成 Agent
  3. Remote MCP 只验证单 server 最小闭环,不做 workflow、多 server 或真实业务数据

因此:

  • Tool / Resource / Prompt 保持各自执行语义
  • reader-skill 只承接本版固定的 local / remote MCP capability
  • project-assistant-service 当前只提供 mock Resource / Prompt / Tool
  • 普通问答主链不因 MCP 扩展退化

开发

安装依赖:

pnpm install

启动开发环境:

pnpm dev

如果需要同时验证 remote MCP 服务:

pnpm dev:pas

上面的命令会同时:

  • 启动 packages/*build:watch
  • 启动 apps/webapp

如果只想单独启动 webapp,可以使用:

pnpm dev:webapp

如果只想单独开启 workspace 包 watch,可以使用:

pnpm build:watch

常用验证

Webapp

pnpm --dir apps/webapp test
pnpm --dir apps/webapp typecheck
pnpm --dir apps/webapp build

Project Assistant Service

pnpm dev:pas
pnpm typecheck:pas
pnpm build:pas

Stream Core

pnpm --dir packages/stream-core test
pnpm --dir packages/stream-core typecheck
pnpm --dir packages/stream-core build

Lint

pnpm lint
pnpm lint:webapp:fix
pnpm lint:packages:fix

版本演进

  • v0.0.4:本地聊天 + Streamdown
  • v0.0.5:最小 Tool Calling
  • v0.0.6:Multi-Tool Runtime
  • v0.0.7:第一层 Skill Runtime(utility-skill
  • v0.0.8reader-skill
  • v0.0.9:MCP Host MVP 接入现有 Skill Runtime
  • v0.0.10:Chat Runtime 收口 + Stream Core package 化

Roadmap

  • 本地聊天闭环
  • Tool Calling
  • Multi-Tool Runtime
  • 第一层 Skill Runtime
  • reader-skill
  • MCP Host MVP
  • Chat Runtime 收口
  • Stream Core package 化
  • 更多 MCP 能力接入
  • Agent Runtime 骨架
  • 持久化与数据层

License

MIT

README mirrored from the source repository 4 months ago. The original is authoritative.

Questions

About AI Mind

How do I install AI Mind?

Run git clone https://github.com/HWYD/ai-mind, then add the server to your MCP client's configuration. Conduid has recorded 0 installs, so the command is known to work with current clients.

Is AI Mind safe to use with an AI agent?

Its trust score is 34 out of 100 (low). Conduid hasn't run static security checks on this repository yet, so review the source yourself before granting it credentials. It has no ConduID identity yet, so agent calls to it are not receipted.

Is AI Mind still maintained?

Conduid hasn't recorded a commit date for this repository yet. Check the repository directly for recent activity.