Vibe Coding & AI IDEs Cheatsheet: .cursorrules, MCP & High-Leverage Prompts

The definitive 2026 reference for Vibe Coding. Production .cursorrules templates, Claude Code workflows, MCP tool definitions, and Ollama local model commands.

1. Production .cursorrules Architecture Template

A focused .cursorrules or CLAUDE.md file prevents hallucination loops and forces the AI agent to follow your exact project conventions.

CODE SNIPPET

# .cursorrules for Modern TypeScript & Next.js
You are a Senior Full-Stack Software Architect.

## Principles
- Always write clean, idiomatic, fully-typed TypeScript 5+.
- Never use 'any'; create explicit discriminated union types.
- Prefer functional composition and small, testable pure functions.
- For state management, prefer server components and URL searchParams before React state.
- Every async function must include robust error handling and structured logging.

## Verification Routine
- Run 'npm run lint' and 'npm run build' before suggesting that a task is complete.
- Only modify the specific files required for the feature.

2. Model Context Protocol (MCP) Python Server Skeleton

FastMCP allows creating custom tools in 10 lines of code that Claude, Cursor, and Windsurf can call directly.

CODE SNIPPET

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("CodebaseAnalyzer")

@mcp.tool()
def inspect_dependency_health(package_name: str) -> str:
    """Analyze security vulnerabilities and license compliance for a package."""
    # Fast tool execution query
    return f"Analysis complete for {package_name}: 0 vulnerabilities detected, MIT licensed."

if __name__ == "__main__":
    mcp.run(transport="stdio")

3. Ollama CLI & Local LLM Management Commands

Ollama enables zero-cost, private local model execution that connects directly to AI IDEs as a custom backend.

CODE SNIPPET

# Run DeepSeek R1 reasoning model locally (8B quantized)
ollama run deepseek-r1:8b

# Run Llama 3.3 for general programming
ollama run llama3.3:70b

# Serve OpenAI-compatible API on http://localhost:11434/v1
ollama serve

# Create a customized model with system prompt
ollama create code-expert -f ./Modelfile

Need another cheat sheet?

We add new reference guides every week based on community requests.

Request a Cheatsheet →