AI-Native Templates MCP Server

Note: This MCP server is optional/advanced. Most users should use the simpler curl-based approach instead.

Do You Need This?

Use curl scripts (recommended for most users):

# Install skills
curl -sL https://raw.githubusercontent.com/shayke-cohen/ai-native-engineer/main/templates/scripts/install-skill.sh | bash -s -- --all

# Install templates
curl -sL https://raw.githubusercontent.com/shayke-cohen/ai-native-engineer/main/templates/scripts/install-templates.sh | bash -s -- --all

Or tell your AI agent:

Read https://raw.githubusercontent.com/shayke-cohen/ai-native-engineer/main/SKILL.md and set up my project

Use this MCP server when you need:

• CI/CD integration with programmatic access
• Custom automation scripts that need template data
• IDE/tool integration beyond Cursor
• Real-time template fetching without shell scripts

Overview

This MCP server is a pure data provider. It fetches templates from GitHub and returns them with llmInstructions that guide the AI agent on how to handle each file. The agent (not the MCP) performs all file operations.

Prerequisites

• Node.js 20+
• gh CLI installed and authenticated:

# Install gh CLI
brew install gh  # macOS
# or see https://cli.github.com/

# Authenticate
gh auth login

Installation

Via npx (recommended)

Add to your .cursor/mcp.json:

{
  "mcpServers": {
    "ai-native-engineer": {
      "command": "npx",
      "args": ["@shaykec/ai-native-engineer"]
    }
  }
}

Local development

cd mcp-server
npm install
npm run build

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "ai-native-engineer": {
      "command": "node",
      "args": ["/path/to/mcp-server/dist/index.js"]
    }
  }
}

Usage

Single Tool: templates

This MCP exposes one tool with two modes:

List Templates

templates({ mode: "list" })

Returns array of available templates with descriptions:

{
  "templates": [
    { "path": "templates/root/AGENTS.md", "destPath": "AGENTS.md", "description": "Core AI agent instructions." },
    { "path": "templates/cursor-rules/security.mdc", "destPath": ".cursor/rules/security.mdc", "description": "Security rules." }
  ]
}

Sync Templates

templates({ mode: "sync" })

Returns all templates with content and handling instructions:

{
  "templates": [
    {
      "destPath": "AGENTS.md",
      "content": "# AGENTS.md - AI Agent Instructions...",
      "llmInstructions": {
        "purpose": "Core AI agent instructions.",
        "action": "merge",
        "preserve": ["Quick Reference Commands", "Project Overview"],
        "update": ["Critical Rules", "Workflows"],
        "hints": "Keep project-specific content, update process sections."
      }
    }
  ],
  "latestCommit": "abc1234",
  "globalInstructions": "For each template..."
}

llmInstructions Actions

| Action | Behavior | |--------|----------| | create-if-missing | Only create file if it doesn't exist | | overwrite | Replace file entirely with template | | merge | Use preserve/update arrays to intelligently merge |

Templates Included

• Root docs: AGENTS.md, CONTRIBUTING.md, SECURITY.md, TESTING.md
• Cursor rules: always.mdc, security.mdc, testing.mdc, code-style.mdc, architecture.mdc
• Scripts: start-issue.sh, finish-issue.sh, check-ci.sh, setup.sh, etc.
• GitHub: PR template, issue templates, CI workflows
• Knowledge: DECISIONS.md, KNOWN_ISSUES.md, LESSONS_LEARNED.md

How It Works

1. Templates are stored in shayke-cohen/ai-native-engineer GitHub repo
2. MCP fetches templates via gh api (uses your existing GitHub auth)
3. Templates are cached for 1 hour
4. Agent receives templates + llmInstructions
5. Agent reads existing files, applies merge logic, writes files
6. No file system access needed by MCP - agent handles everything

Agent Workflow

When a user asks to sync templates:

1. Agent calls templates({ mode: "sync" })
2. For each template in response:
   • Read existing file at destPath (if any)
   • Apply action from llmInstructions
   • Write file using agent's file tools
3. Report: X created, Y updated, Z skipped
4. Remind user: chmod +x scripts/*.sh

MCP vs Curl Scripts Comparison

| Feature | Curl Scripts | MCP Server | |---------|-------------|------------| | Setup | Zero setup | Requires npm + gh CLI | | Authentication | None needed | Needs gh auth login | | Use in Cursor | Agent reads SKILL.md | Agent calls MCP tool | | Offline use | Downloads files | Requires live connection | | CI/CD friendly | Yes (bash) | Yes (programmatic API) | | Update checking | Via --check-updates | Returns latestCommit | | Smart merge | Downloads as .new | Returns llmInstructions |

Recommendation: Start with curl scripts. Only add MCP if you need programmatic access.

Source Repository

Templates: https://github.com/shayke-cohen/ai-native-engineer

Related

• ai-tester - AI-powered testing MCP server