One kit. All your AI coding tools. Agent Skills management, MCP configuration sync, and iOS engineering rules — unified for Cursor, CodeBuddy, Codex, Claude Code, Gemini CLI, Continue, Cline, Qwen Code, and Xcode Coding Assistant.
ai-coding-kit is a local-first AI coding workflow toolkit. Define your MCP servers, API keys, Agent Skills, and platform settings once — auto-sync to every AI coding host you use.
面向 AI Coding / Agentic Coding / MCP（模型上下文协议）的多工具本地工程化工具包。为同时使用多个 AI 编码工具的开发者提供统一的 Agent Skill 维护、MCP 配置同步、iOS 工程规则和本地嵌入式知识库。
git clone https://github.com/i-stack/ai-coding-kit.git
cd ai-coding-kit
# 初始化本地配置（从 env/*.example 模板创建缺失文件，幂等）
bash install.sh
# 唯一需要编辑的文件：填入真实 API Keys / Tokens
$EDITOR env/secrets.json
# 一键同步
bash sync.sh当前主要开发和测试环境为 macOS。部分平台模块（如 Codex 同步中的 Xcode 集成、.zshrc 导出）在 macOS 外不可用。
欢迎 Windows 用户在 Windows 上验证并提交 PR。核心同步逻辑已尽量保持跨平台，适配改动预计较小。
🔒 纯本地同步，API 永不离机。 sync 引擎仅在你的本机文件系统内工作——将
env/secrets.json中的密钥注入 MCP 定义，渲染到各平台本地配置文件。不会上传任何数据到外部服务器，不会调用任何网络 API。 你的 API Key 始终只保存在这台机器上。可查看 sync 源码 。🔒 Local-only sync. Your API keys never leave this machine. The sync engine works entirely within your local filesystem — it reads secrets from
env/secrets.json, injects them into MCP definitions, and renders them into each platform's local config files. No data is uploaded to any external server. No network API is called. Your API keys stay on this machine, always. verify in the sync source code.
各模块有独立的 README，按需深入：
| 模块 | 说明 | 文档 |
|---|---|---|
| skills-engineering/ | Agent Skill 内容源、多端同步、受控演进 | README |
| sync/ | MCP 配置同步引擎，注入 secrets 渲染到各平台原生格式 | README |
| env/ | 配置数据源（secrets + MCP 定义 + 平台配置） | README |
| skills-engineering/plan-reviews/ | Plan 知识库 —— 自动索引 plan-grill/cross-model-review 产物，语义搜索 + 实体图谱 | README |
| hooks/ | 项目钩子脚本（xmcp 初始化等） | README |
| .githooks/ | Git 提交/推送守卫（pre-commit + pre-push） | README |
| Tool | Synced Config |
|---|---|
| Cursor | .cursor/mcp.json |
| CodeBuddy | .codebuddy/mcp.json, models.json, skills/ |
| Claude Code | .claude.json, settings.json, skills/ |
| Codex CLI | .codex/config.toml |
| Gemini CLI | Environment variables |
| Continue | .continue/config.yaml |
| Cline (VSCode) | MCP settings JSON, skills/ |
| Qwen Code | settings.json env, skills/ |
| Xcode Coding Assistant | Codex + Claude Agent config paths |
bash install-hooks.sh启用 pre-commit（规则变更治理）、post-commit（evolution 历史自动 GC）和 pre-push（推送前强制同步校验）。详见 .githooks/README.md。
Model Context Protocol (MCP) is an open protocol that lets AI coding tools connect to external services — GitHub, Playwright, databases, APIs, design tools — through a standardized interface. ai-coding-kit gives you one place to define all your MCP servers and syncs them to every tool that supports MCP.
- Developers using multiple AI coding tools — define MCP servers and Agent Skills once, sync everywhere.
- iOS / Swift engineers — production-grade AI coding rules for Swift, SwiftUI, UIKit, concurrency, testing, and migration.
- AI infrastructure builders — local embedded knowledge base with semantic search and entity graph, zero external services.
- Team maintainers — single source of truth for MCP configuration, API keys, and model settings.