@ai-agencee/mcp

MCP (Model Context Protocol) Server for @ai-agencee/cli. This server enables AI assistants to automatically load and enforce your project's ULTRA_HIGH coding standards when working with your codebase.

Features

✅ Initialize AI Sessions - Load TADEO/ULTRA_HIGH standards automatically
✅ Project Validation - Check project structure against rules
✅ Standards Discovery - Access project rules, patterns, and guidelines
✅ Bootstrap Resources - Load AI session initialization config

Installation

npm install @ai-agencee/mcp

Quick Start

1. Start the MCP Server

npx @ai-agencee/mcp

The server starts on stdio and connects to your MCP client.

2. Configure Your AI Assistant

For Claude or other MCP-compatible AI assistants, add this server to your configuration:

Claude Desktop (macOS/Windows/Linux)

Edit ~/.config/Claude/claude_desktop_config.json (or %APPDATA%\Claude\claude_desktop_config.json on Windows):

{
  "mcpServers": {
    "ai-kit": {
      "command": "npx",
      "args": ["@ai-agencee/mcp"]
    }
  }
}

3. Use in AI Sessions

Once configured, use these tools in your AI assistant:

@init              # Initialize session with ULTRA_HIGH standards
@check             # Validate project structure
@rules             # View coding standards
@patterns          # View design patterns
@bootstrap         # Get setup instructions

Example prompt:

@init
@rules

Based on the project rules, how should I structure my TypeScript component?

Tools Available

@init

Initializes AI session with TADEO/ULTRA_HIGH standards.

Parameters:

• strict (boolean, optional): Enable STRICT_MODE. Default: true

Returns:

• Configuration summary
• Project rules loaded
• Available tools list
• Next steps

@check

Validates current project structure against rules.

Returns:

• Type safety check status
• Linting results
• Testing coverage
• Rules compliance

@rules

Get your project's coding standards and conventions.

Parameters:

• format (string): markdown or text. Default: markdown

Returns:

• File naming conventions
• Export patterns
• Type requirements
• Testing standards
• Forbidden patterns

@patterns

Get design patterns and architecture guidelines for your project.

Parameters:

• format (string): markdown or text. Default: markdown

Returns:

• Architecture patterns
• Best practices
• Common examples
• Anti-patterns

@bootstrap

Get bootstrap configuration and setup instructions.

Parameters:

• format (string): markdown, text, or config. Default: markdown

Returns:

• Project setup guide
• Environment configuration
• Development workflow

Resources Available

MCP Resources provide direct access to project files:

• bootstrap://init - AI session initialization guide
• bootstrap://rules - Project coding rules
• bootstrap://patterns - Design patterns guide
• bootstrap://manifest - Project manifest and capabilities

ULTRA_HIGH Standards Reference

The MCP server enforces these TADEO/ULTRA_HIGH standards:

| Standard | Rule | |----------|------| | Naming | kebab-case (no camelCase) | | Files | One export per file | | Types | No any type allowed | | Functions | export const Name = function(...) { ... } | | Classes | FORBIDDEN | | Testing | 95% minimum coverage required | | Performance | ≤10% solid-js overhead |

Configuration Flags

U=TADEO                 # Universe: TADEO framework
STD=ULTRA_HIGH          # Standard level
COM=BRUTAL              # Communication: Direct and honest
VERBOSITY=0             # Minimal output
POLITE=0                # Direct without pleasantries
PROSE=0                 # Code-focused, minimal narrative
HEADLESS=1              # Non-interactive mode
DELEGATE=0              # No delegating tasks
STRICT_MODE=1           # Strict type and rule checking
IGNORE_HISTORY=1        # Fresh context each session
NO_CHAT=1               # No chat mode, direct instructions

Architecture

The MCP server is built on the @modelcontextprotocol/sdk with:

• Transport: Stdio-based communication with MCP clients
• Tools: Four request handlers for AI assistant integration
• Resources: File-based resources for project documentation
• Error Handling: Graceful error messages and fallback behavior

Integration Examples

With Claude

👤 User: @init

🤖 Assistant: 
# AI SESSION INITIALIZED

## Configuration
U=TADEO
STD=ULTRA_HIGH
...

Ready to start development with strict standards applied!

---

👤 User: @rules

How should I name my variables?

🤖 Assistant:
All variables must use kebab-case naming:
✓ user-input
✓ api-response  
✗ userData
✗ apiResponse

With Other AI Assistants

Any MCP-compatible client can use this server:

• Anthropic Claude
• OpenAI Assistants (with MCP wrapper)
• Cline/cursor
• VS Code Copilot (with MCP integration)

Troubleshooting

Server not starting

Ensure you have Node.js 18+ installed:

node --version  # Should be v18.0.0 or higher

Cannot find project files

The server reads files from the current working directory. Make sure you're running it from your project root:

cd /path/to/project
npx @ai-agencee/mcp

File not found errors

The server looks for these paths relative to project root:

• .github/copilot-instructions.md
• src/.ai/rules.md
• src/.ai/patterns.md
• .github/ai/manifest.xml

Create these files using @ai-agencee/cli:

npx @ai-agencee/cli init --strict

License

MIT - See LICENSE for details

Related Packages

• @ai-agencee/cli - Command-line scaffolding tool
• @ai-agencee/core - Shared utilities and templates