AI Integration

Kruda includes a built-in MCP (Model Context Protocol) server that enables AI coding assistants to scaffold, generate code, and analyze your Kruda project.

Setup

Install CLI

go install github.com/go-kruda/kruda/cmd/kruda@latest

New Projects

kruda new includes .mcp.json automatically:

kruda new myapp
cd myapp
# Open in Claude Code or Cursor — MCP tools ready to use

Existing Projects

kruda mcp init        # generates .mcp.json + .cursor/mcp.json
kruda mcp --test      # verify it works

Available Tools

Tool	Description
kruda_new	Scaffold a new project (minimal, api, fullstack)
kruda_add_handler	Generate a route handler with optional typed C[T] pattern
kruda_add_resource	Generate a CRUD ResourceService[T, string] implementation
kruda_list_routes	Scan Go source files and list all registered routes
kruda_suggest_wing	Suggest Wing Feather hints for routes
kruda_docs	Look up Kruda docs and code examples by topic

How It Works

The MCP server runs locally on your machine as a stdio process. Your AI assistant communicates with it via JSON-RPC:

AI Assistant (Claude Code / Cursor)
       ↓ stdio JSON-RPC
   kruda mcp
       ↓
   reads/writes files in your project

No data leaves your machine. No remote server required.

Configuration

.mcp.json (Claude Code):

{
  "mcpServers": {
    "kruda": {
      "command": "kruda",
      "args": ["mcp"]
    }
  }
}

.cursor/mcp.json (Cursor): same format.

Why AI-Friendly?

Kruda's typed API makes AI code generation more reliable:

• Typed handlers — AI generates a struct, gets compile-time validation for free
• Auto CRUD — AI says "create CRUD for Product" → one line of code
• Struct tags — validation rules are explicit, not hidden in handler logic
• 21 examples — AI reads patterns and generates consistent code