Quilt

Quilt is an open-source deployment system that transforms websites and code repositories into Claude Desktop MCP-native data sources.
It ingests repos, extracts tagged content, computes semantic embeddings using Cohere, and publishes them to a searchable API so LLMs can query them instantly—without building custom APIs.

---

Quilt Overview

Quilt reduces the gap between human-facing content and machine-facing interfaces by:

• Making websites MCP-native: Any GitHub repo can be deployed and indexed for Claude Desktop search.
• Eliminating manual indexing: Content is automatically extracted and vectorized on deployment.
• Precomputing embeddings: Cohere embeddings are generated ahead of time for instant semantic queries.
• GitHub OAuth integration: Simple authentication and repository selection through GitHub.

Think: "Don't just deploy to the web—deploy to LLMs with Quilt."

---

Quilt Advantages

Quilt creates benefits for three groups at once:

• For LLMs & Claude Desktop → Faster and more efficient retrieval. Pre-indexed vectors mean no wasted compute or latency at query time.
• For Content Owners → A simple way to make their GitHub repositories instantly accessible to Claude Desktop. Just tag content with data-llm attributes and deploy.
• For Developers → No need to build or maintain custom search APIs. Quilt surfaces data through a unified interface automatically.

Additional benefits:

• GitHub-native workflow → Deploy directly from your existing repositories.
• Cohere embeddings → State-of-the-art semantic search using embed-english-v3.0.
• PostgreSQL storage → Scalable vector database with full-text search capabilities.
• Real-time deployment → Push changes and see them reflected in search immediately.
• Web dashboard → Visual interface for managing deployments and viewing statistics.

---

How Quilt Works

Quilt's architecture combines modern NLP with GitHub-native developer workflows:

1. GitHub Authentication – Users authenticate via GitHub OAuth and select repositories.
2. Repository Deployment – Selected repos are fetched and processed via GitHub API.
3. Content Parsing – HTML/JSX/React files are scanned for data-llm attributes.
4. Content Extraction – Elements with data-llm tags have their text content extracted.
5. Vectorization – Cohere's embed-english-v3.0 generates semantic embeddings.
6. Database Storage – Content, embeddings, and metadata stored in PostgreSQL.
7. Search API – RESTful endpoints provide vector, keyword, and hybrid search.
8. MCP Integration – Claude Desktop can search through deployed content via MCP server.

Tech stack:

• Frontend → Next.js with React and Tailwind CSS
• Backend → FastAPI with Python
• Database → PostgreSQL with vector search capabilities
• Embeddings → Cohere API (embed-english-v3.0)
• Deployment → Render (backend) + Vercel (frontend)

---

Quilt Deployment Flow

1. GitHub OAuth – Authenticate and authorize Quilt to access repositories.
2. Repository Selection – Choose which repos to deploy from your GitHub account.
3. Content Processing – Quilt fetches repo files and scans for data-llm attributes.
4. Text Extraction – Tagged content is extracted and cleaned.
5. Embedding Generation – Cohere API generates semantic vectors for each content section.
6. Database Indexing – Content, embeddings, and metadata stored in PostgreSQL.
7. Search Availability – Content becomes searchable via API and Claude Desktop MCP.
8. Dashboard Monitoring – View deployment status and search statistics in web interface.

---

Quilt Content Annotation

Use data-llm attributes in your HTML/JSX to define searchable regions:

• data-llm="title" → For titles/headings
• data-llm="content" → For descriptive sections
• data-llm="description" → For detailed explanations

Example (React component):

export default function FeatureSection() {
  return (
    <div>
      <h2 data-llm="title">Vector Search Technology</h2>
      <p data-llm="content">
        Quilt uses Cohere embeddings to enable semantic search across 
        your deployed repositories, making content instantly discoverable.
      </p>
      <div data-llm="description">
        The system processes HTML and JSX files, extracting tagged content
        and generating high-quality vector representations for Claude Desktop.
      </div>
    </div>
  )
}

Example (HTML file):

<h1 data-llm="title">Getting Started with Quilt</h1>
<p data-llm="content">Deploy your repositories to make them searchable by Claude Desktop.</p>
<section data-llm="description">
  Quilt automatically processes your content and creates semantic embeddings,
  enabling powerful search capabilities for AI assistants.
</section>

---

API Endpoints

Quilt provides several REST endpoints for search and management:

• POST /deploy → Deploy a GitHub repository
• GET /search → Search all deployed content
• GET /search/{user_id} → Search specific user's content
• GET /deployments/{user_id} → List user's deployments
• GET /stats → Get database statistics
• DELETE /deployments/{deployment_id} → Remove a deployment

---

Getting Started

1. Visit the Quilt web interface
2. Authenticate with GitHub using OAuth
3. Select repositories you want to make searchable
4. Deploy with one click
5. Search your content via the web interface or Claude Desktop

The deployed content becomes immediately available for semantic search through both the web API and Claude Desktop's MCP integration.