Craig

A Model Context Protocol (MCP) implementation for searching markdown documentation from UnJS and Nuxt ecosystems.

Overview

Craig provides a specialized MCP service for retrieving, indexing, and searching through markdown documentation from repositories within the UnJS and Nuxt organizations. It enables AI assistants to access up-to-date information about these libraries and frameworks.

Features

• Automatically fetches markdown documentation from UnJS and Nuxt repositories
• Stores files in Cloudflare R2 storage using AutoRAG
• Provides MCP-compatible API endpoints for AI tools
• Scheduled tasks for regular content updates

Technical Stack

• Built with Nuxt
• Uses NuxtHub for AI and Workers
• Deploys to Cloudflare Workers
• Leverages Cloudflare AutoRAG for retrieval-augmented generation
• Implements Model Context Protocol for AI integration

API Endpoints

• /api/blobs - List all stored content
• /files/:pathname - Serve specific files
• /sse - Server-Sent Events endpoint with MCP integration for AI search

Development

# Install dependencies
pnpm install

# Start development server
pnpm dev

# Build for production
pnpm build

Indexed Repositories

Currently indexed repositories include:

• unjs/h3
• unjs/db0
• unjs/ungh
• unjs/jiti
• unjs/ofetch
• unjs/consola
• unjs/unstorage

To modify the tracked repositories, edit server/utils/repos.ts.

Roadmap

• Support for additional repositories from UnJS, Nuxt, and related ecosystems
• Indexing GitHub issues for context on bug reports and feature requests
• Integration with GitHub discussions for design decisions and community insights
• Tracking pull requests to provide information on upcoming features and fixes
• Support for additional file formats beyond markdown
• Advanced search filtering and categorization
• API improvements for more specialized queries

---

Created with ❤️ for UnJS and Nuxt ecosystems.