AI Code Toolkit

Scale your AI coding agents with scaffolding, architecture patterns, and validation rules

A collection of Model Context Protocol (MCP) servers and tools that help AI coding agents maintain consistency, follow conventions, and scale with your codebase.

Contents

• Why This Exists

• Core Pillars

• Our Approach

• Getting Started

  • Prerequisites

  • Quick Start

• Project Structure Support

  • Monorepo (Default)

  • Monolith

  • Creating Projects

  • Adding Features

• Development Workflow

• Packages

• Supported Integrations

  • Coding Agents

  • Spec-Driven Development Tools

• Contributing

• License

Why This Exists

As projects evolve from MVP to production, they develop patterns, conventions, and opinionated approaches. Custom instructions alone struggle to ensure AI agents follow these requirements—especially as complexity grows and context windows fill up.

AI Code Toolkit provides building blocks to scale coding agent capabilities:

• Generate code that follows your team's conventions

• Enforce architectural patterns automatically

• Validate agent outputs programmatically

• Work with any AI coding agent (Claude, Cursor, etc.)

• Support any tech stack (Next.js, React, or custom frameworks)

The Perfect Complement to Spec Driven Development

While Spec Driven Development tells AI agents what to build (requirements, features, behavior), AI Code Toolkit tells them how to build it (patterns, conventions, validation). Together, they create a complete development workflow:

• Specs → Define product requirements and business logic

• AI Code Toolkit → Enforce technical standards and architectural consistency

Whether you're bootstrapping a new project or managing a complex monorepo, these tools ensure AI agents integrate seamlessly with your development workflow.

Core Pillars

1. Scaffolding Templates

Combine templating with LLMs to generate standardized code that follows your internal conventions while reducing maintenance overhead.

2. Architecture + Design Patterns

Convention over configuration scales. Like Ruby on Rails or Angular, opinionated approaches make code predictable—for both humans and AI agents.

3. Rules

Pre-flight guidance + post-flight validation = consistent output. Rules provide programmatic checks (quantitative or qualitative) to enforce your processes.

Our Approach

Agent Agnostic

Works with any AI coding agent (Claude Code, Cursor, Windsurf, etc.). Each library provides:

• MCP tools for integration with MCP-compatible agents

• CLI commands for scripting deterministic workflows

Tech Stack Agnostic

Built-in templates for popular frameworks:

• Next.js 15

• TypeScript Libraries & MCP Packages

• More coming soon (Vite + React, Remix, etc.)

Don't see your stack? Use the built-in MCP tools to generate custom templates—the system is fully extensible.

Coding Tool Specific

Maximize effectiveness by combining three layers:

1. MCP Servers → Let tools guide the agent with their default prompts

2. Custom Instructions → Use CLAUDE.md, AGENTS.md to specify when to use MCP tools

3. Hooks → Intercept tool calls to enforce workflows (e.g., require scaffolding for new files)

Experiment with these layers to find the right balance for your project. There's no one-size-fits-all solution.

Getting Started

Prerequisites

• Node.js: >= 18 (LTS recommended)

• Package Manager: pnpm (or npm/yarn)

• Git: >= 2.13.2

Quick Start

Get started with AI Code Toolkit in 3 simple steps:

Step 1: Initialize Templates (or Create New Project)

For Existing Projects:

Run the init command to download official templates:

This automatically downloads official templates (Next.js 15, TypeScript libraries, MCP packages) to your workspace.

For New Projects:

Run init without a git repository and it will guide you through creating a new project:

The command will prompt you for project details, initialize git, and download templates.

Step 2: Configure MCP Servers

Add the MCP servers to your AI coding tool's configuration:

For Claude Code, Cursor, Windsurf, or any MCP-compatible tool:

Add to your MCP settings (e.g., claude_desktop_config.json, .cursor/mcp.json):

Note for Claude Code users: You can also use the plugin marketplace for automatic configuration.

Step 3: Use MCP Tools and Slash Commands

Once configured, your AI agent has access to:

Scaffolding Tools (scaffold-mcp):

• list-boilerplates - Show available project templates

• use-boilerplate - Create new project from template

• list-scaffolding-methods - Show features you can add to existing projects

• use-scaffold-method - Add features (pages, components, API routes, etc.)

Architecture Tools (architect-mcp):

• get-file-design-pattern - Get design patterns for a specific file

• review-code-change - Review code against rules and standards

Slash Commands (for compatible agents like Claude Code):

• /scaffold-mcp:scaffold-application - Create a new application

• /scaffold-mcp:scaffold-feature - Add features to existing projects

• /scaffold-mcp:generate-boilerplate - Create custom boilerplate templates

• /scaffold-mcp:generate-feature-scaffold - Create custom feature scaffolds

Available Templates:

• nextjs-15-drizzle: Next.js 15 + App Router + TypeScript + Tailwind CSS 4 + Storybook + Optional Drizzle ORM

• typescript-lib: TypeScript library with ESM/CJS builds, testing, and documentation

• typescript-mcp-package: Model Context Protocol (MCP) server package template

Project Structure Support

AI Code Toolkit supports both monorepo and monolith project architectures. The tools automatically detect your project type and adapt accordingly.

Monorepo (Default)

Multi-project workspaces using tools like Nx, Lerna, or Turborepo:

Configuration: Each project has project.json with sourceTemplate field

Monolith (New!)

Single-application codebases:

Configuration: toolkit.yaml at workspace root with projectType and sourceTemplate

Alternative: Configure via package.json:

Creating Projects

New Project with Interactive Setup (recommended):

New Project with CLI Options (non-interactive):

Existing Workspace (just initialize templates):

Adding Features

Use scaffold-mcp MCP tools to add features. The tools automatically detect your project type:

MCP Tools:

• list-scaffolding-methods - See available features for your project

• use-scaffold-method - Add pages, components, routes, etc.

CLI Alternative: For manual usage, see the @agiflowai/aicode-toolkit CLI documentation.

Key Points:

• Templates are architecture-agnostic (same templates work for both)

• Tools auto-detect project type from configuration files

• Use --project-type monolith option when creating new projects with init command

• Configuration priority: project.json → toolkit.yaml → package.json

Development Workflow

The AI Code Toolkit packages work together to create a complete development workflow for AI coding agents. Here's how they integrate:

Complete Workflow: From Project Creation to Code Review

How They Work Together

scaffold-mcp and architect-mcp are complementary:

Integration Points:

1. Shared Templates: Both use the same template structure

   templates/nextjs-15/ ├── scaffold.yaml ← scaffold-mcp: Defines boilerplates/features ├── architect.yaml ← architect-mcp: Defines design patterns └── RULES.yaml ← architect-mcp: Defines coding rules

2. Project Configuration: Projects reference templates via project.json

   { "name": "my-app", "sourceTemplate": "nextjs-15" }

3. Workflow Stages:

   • Pre-coding: scaffold-mcp generates boilerplate → architect-mcp shows patterns

   • During coding: architect-mcp provides guidance → AI agent writes code

   • Post-coding: architect-mcp reviews code → Identifies violations

   • Iteration: scaffold-mcp adds features → architect-mcp validates

Example: Building a Next.js Application

Step 1: Create Project AI agent uses scaffold-mcp MCP tool use-boilerplate to create the application.

Result: Project created with Next.js structure, architect.yaml, and RULES.yaml

Step 2: Understand Patterns (Before writing custom code) AI agent uses architect-mcp MCP tool get-file-design-pattern to understand patterns.

Result: Shows "Next.js App Router Pattern" and applicable rules

Step 3: Add Feature (Standard features) AI agent uses scaffold-mcp MCP tool use-scaffold-method to add a route.

Result: Route created following template patterns

Step 4: Write Custom Code (AI agent writes business logic)

Step 5: Review Code AI agent uses architect-mcp MCP tool review-code-change to validate the code.

Result: Validates against RULES.yaml (named exports, error handling, etc.)

Why This Approach Works

1. Templates as Single Source of Truth: Both tools read from same template definitions

2. Separation of Concerns:

   • scaffold-mcp: Generates repetitive code

   • architect-mcp: Guides unique code

3. Progressive Enhancement:

   • Start with scaffolding (fast, consistent)

   • Add custom logic (AI-assisted, pattern-guided)

   • Validate output (automated review)

4. Feedback Loop: Reviews inform future scaffolding and patterns

Packages

@agiflowai/aicode-toolkit

Unified CLI for initializing projects, managing templates, and configuring MCP servers. Provides interactive workflows for project setup with automatic detection of coding agents (Claude Code, Cursor, Gemini CLI, Codex CLI, GitHub Copilot).

Key Features:

• Interactive project initialization (monorepo or monolith)

• Template management and download from GitHub

• Automatic MCP server configuration

• Coding agent auto-detection

• New project scaffolding with git setup

View full documentation →

@agiflowai/one-mcp

Unified MCP proxy server that enables progressive tool discovery to dramatically reduce token usage. Instead of loading all tools from all servers upfront (40,000+ tokens), agents load only 2 meta-tools initially (~400 tokens) and discover tools on-demand.

Key Features:

• Progressive tool discovery (90%+ token reduction)

• Smart proxy for multiple MCP servers

• On-demand tool loading and routing

• Flexible configuration (local/remote YAML/JSON)

• Environment variable interpolation

• Support for stdio, HTTP, and SSE transports

View full documentation →

@agiflowai/scaffold-mcp

MCP server for scaffolding applications with boilerplate templates and feature generators. Works with any MCP-compatible coding agent.

Key Features:

• Create projects from boilerplate templates

• Add features to existing projects (pages, components, services)

• Template management (initialize, add from repositories)

• Built-in templates: Next.js 15, TypeScript libraries, MCP packages

• Multiple transport modes: stdio, HTTP, SSE

• Standalone CLI and slash command prompts

View full documentation →

@agiflowai/architect-mcp

MCP server for architecture design, code quality enforcement, and design pattern guidance. Validates code against project-specific rules and patterns.

Key Features:

• Design pattern guidance for specific files

• Code review against template-specific rules

• Architecture patterns (architect.yaml)

• Coding standards and rules (RULES.yaml)

• Optional LLM-powered analysis

• Multiple transport modes: stdio, HTTP, SSE

View full documentation →

@agiflowai/coding-agent-bridge

Service layer for integrating with AI coding agents. Provides a unified interface for detecting, configuring, and invoking different coding assistants programmatically.

Key Features:

• Automatic coding agent detection (Claude Code, Cursor, Gemini CLI, etc.)

• Unified interface for MCP configuration and custom instructions

• LLM invocation API for using coding agents as pure LLMs

• Agent-specific adapters with consistent API

• Support for stdio, HTTP, and SSE transports

Currently Supported Coding Agents:

• Claude Code - Full support with .mcp.json configuration and CLAUDE.md/AGENTS.md instructions

• Cursor - MCP configuration and .cursorrules instructions

• Gemini CLI - .gemini workspace detection and configuration

• Codex CLI - .codex workspace detection and configuration

• GitHub Copilot - MCP support and .github/copilot-instructions.md

Coming Soon:

• Windsurf

• Zed

• VS Code with Continue/Cody extensions

View full documentation →

Supported Integrations

Coding Agents

AI Code Toolkit integrates with popular AI coding assistants through the coding-agent-bridge package:

Features:

• Auto-detection: Automatically detects installed coding agents in your workspace

• Unified API: Same interface for all agents (configuration, instructions, invocation)

• MCP Configuration: Programmatic setup of MCP servers

• LLM Invocation: Use coding agents as pure LLMs (no tool use)

Spec-Driven Development Tools

AI Code Toolkit supports spec-driven development through the aicode-toolkit CLI:

How It Works:

Contributing

We welcome contributions! Whether it's bug reports, feature requests, or pull requests—all contributions are appreciated.

How to contribute:

1. Fork the repository

2. Create a feature branch (git checkout -b feature/amazing-feature)

3. Make your changes

4. Run tests and linting (pnpm test && pnpm lint)

5. Commit your changes (follow conventional commits)

6. Push to your branch (git push origin feature/amazing-feature)

7. Open a Pull Request

See CONTRIBUTING.md for detailed guidelines.

License

AGPL-3.0 © AgiflowIO

Built with ❤️ by the AgiflowIO team

• Report Issues

• Discussions

• Discord Community

• Website