swift-patterns-mcp

An MCP server providing curated Swift and SwiftUI best practices from leading iOS developers — with intelligent search, persistent memory, and optional premium integrations.

Want an Agent Skill?

If you want a lightweight, portable Swift/SwiftUI best-practices package without runtime tooling, check out:

swift-patterns-skill: Designed as a portable Agent Skill focused on Swift/SwiftUI patterns, architecture guidance, and decision-making frameworks.

Key difference:

• swift-patterns-skill = Static guidance (portable, no runtime)
• swift-patterns-mcp = Dynamic tooling (search, retrieval, premium features)

Note: This repo is an MCP server only. It does not ship an Agent Skill (SKILL.md) or skill references.

What does this MCP provide?

swift-patterns-mcp delivers runtime tools for accessing Swift/SwiftUI best practices:

• 🔎 Search & retrieval across curated sources
• 🧠 Persistent memory with cross-session recall
• 🔄 Auto-refreshing content from RSS feeds and GitHub
• 🎯 Intelligent filtering by quality and relevance
• 🔐 Premium integrations (optional Patreon support)

Ideal for:

• Active Development: "How do I implement pull-to-refresh in SwiftUI?" answered instantly without leaving your IDE
• Architecture Decisions: Compare MVVM vs. TCA patterns with concrete examples from trusted sources
• Staying Current: Access the latest patterns and best practices as they're published by leading iOS developers
• Team Standards: Build a searchable reference of approved patterns for your organization
• AI-Powered Workflows: Enable agents to query "Show me Sundell's approach to dependency injection" with consistent, quality responses

🌟 Features

• 🎓 Expert Knowledge Base: Patterns from Swift by Sundell, Antoine van der Lee, Nil Coalescing, and more
• 🔍 Intelligent Search: Query by topic, pattern, or specific iOS concept
• 💾 Persistent Memory: Cross-session recall with Memvid storage
• 🧠 Semantic Search: Optional AI-powered fallback for better conceptual matches
• 📚 Multiple Sources: Aggregates knowledge from trusted educators
• 🔄 Auto-Updates: Content refreshes automatically from RSS feeds
• ⚡ Fast Performance: Efficient caching and indexed search

Content Sources

Free Sources

These sources are publicly available but benefit from MCP's fetching, caching, and search capabilities:

Premium Sources

Premium content requires OAuth authentication and active subscriptions:

Access exclusive content from top iOS educators: Kavsoft, SwiftUI Codes, sucodee and many more. Get tutorials, code samples, and expert guidance directly from creators you support.

📋 Prerequisites

• Node.js 18.0.0 or higher
• MCP-Compatible AI Assistant: Claude Desktop, Cursor, Windsurf, VS Code with Copilot, or Claude Code

🚀 Quick Start

Run Setup

npx -y swift-patterns-mcp@latest

In an interactive terminal, this opens the setup wizard.
When launched by an MCP client (non-interactive stdio), it runs as the MCP server automatically.

Interactive Setup Wizard

npx -y swift-patterns-mcp@latest setup

If installed globally, you can also run:

swift-patterns-mcp setup

The wizard helps you choose:

• Config scope (local project vs global)
• MCP client (Cursor, Claude Code, Windsurf, VS Code)
• Optional Patreon setup prompt

Non-interactive Setup (CI/Scripts)

# Cursor
npx -y swift-patterns-mcp@latest setup --cursor --global
npx -y swift-patterns-mcp@latest setup --cursor --local

# Claude Code
npx -y swift-patterns-mcp@latest setup --claude --global

# Windsurf
npx -y swift-patterns-mcp@latest setup --windsurf --global

# VS Code
npx -y swift-patterns-mcp@latest setup --vscode --local

# All clients
npx -y swift-patterns-mcp@latest setup --all --global

Use --global (-g) or --local (-l) to skip the location prompt.
Use --cursor, --claude, --windsurf, --vscode, or --all to skip the client prompt.

Configure Your AI Assistant

Cursor

Or manually add to Cursor Settings → Tools → MCP Servers:

.cursor/mcp.json:

{
  "mcpServers": {
    "swift-patterns": {
      "command": "npx",
      "args": ["-y", "swift-patterns-mcp@latest"]
    }
  }
}

Alternatively, add to ~/.cursor/mcp.json. See Cursor documentation for details.

Claude Code

Run in your terminal:

claude mcp add swift-patterns -- npx -y swift-patterns-mcp@latest

Or manually add to .mcp.json:

{
  "mcpServers": {
    "swift-patterns": {
      "command": "npx",
      "args": ["-y", "swift-patterns-mcp@latest"]
    }
  }
}

Restart Claude Code and run /mcp to verify. See Claude Code MCP documentation for details.

Windsurf

Add to .windsurf/mcp.json:

{
  "mcpServers": {
    "swift-patterns": {
      "command": "npx",
      "args": ["-y", "swift-patterns-mcp@latest"]
    }
  }
}

Restart Windsurf to activate. See Windsurf MCP documentation for details.

VS Code

Add to .vscode/mcp.json:

{
  "mcp": {
    "servers": {
      "swift-patterns": {
        "command": "npx",
        "args": ["-y", "swift-patterns-mcp@latest"]
      }
    }
  }
}

Open .vscode/mcp.json and click Start next to the swift-patterns server. See VS Code MCP documentation for details.

Test It Out

Try these queries:

"Show me SwiftUI animation patterns"
"What does Sundell say about testing?"
"Explain navigation patterns in SwiftUI"

🔧 Configuration

Configuration is automatically created at ~/.swift-patterns-mcp/config.json:

{
  "sources": {
    "sundell": { "enabled": true },
    "vanderlee": { "enabled": true },
    "nilcoalescing": { "enabled": true },
    "pointfree": { "enabled": true },
    "patreon": { "enabled": false, "configured": false }
  },
  "prefetchSources": true,
  "semanticRecall": {
    "enabled": false,
    "minLexicalScore": 0.35,
    "minRelevanceScore": 70
  },
  "memvid": {
    "enabled": true,
    "autoStore": true,
    "useEmbeddings": false,
    "embeddingModel": "bge-small"
  }
}

Note: configured only applies to premium sources. Free sources are treated as configured by default.

Persistent Memory with Memvid

Memvid provides persistent semantic memory that improves recall across sessions. Unlike in-memory caching, Memvid stores patterns in a single-file database that persists between server restarts.

Features:

• 💾 Persistent Storage: Patterns stored in ~/.swift-patterns-mcp/swift-patterns-memory.mv2
• 🔁 Cross-Session Recall: Find patterns from previous searches after server restart
• 🧠 Semantic Search: Optional embedding-based similarity search
• 🚀 Automatic Storage: Patterns stored during searches
• ⚡ Fast Retrieval: Built-in BM25 + optional vector search

Configuration:

{
  "memvid": {
    "enabled": true,              // Enable Memvid persistent memory
    "autoStore": true,            // Automatically store patterns during searches
    "useEmbeddings": false,       // Use semantic embeddings (requires model download)
    "embeddingModel": "bge-small" // Options: "bge-small", "openai-small"
  }
}

When to enable:

• You want patterns to persist across server restarts
• You frequently search for similar topics
• You need cross-session semantic memory

Note: Memvid complements MiniSearch (fast in-session search) and semantic recall (in-session fallback). All three work together:

1. MiniSearch: Fast lexical search within current session
2. Semantic recall: Activates for poor lexical results (in-session)
3. Memvid: Cross-session persistent memory and recall

Semantic Recall (Optional AI Enhancement)

Semantic recall provides AI-powered semantic search as a fallback when keyword search returns poor results. It uses transformer embeddings to understand query intent and find conceptually similar patterns.

Features:

• 🧠 Automatically activates when keyword search scores are low
• 🎯 Uses sentence transformers to understand meaning beyond keywords
• 📊 Quality filtering to index only high-relevance patterns
• ⚡ Efficient embedding caching

Configuration:

{
  "semanticRecall": {
    "enabled": false,              // Enable semantic recall
    "minLexicalScore": 0.35,       // Activate when keyword search < 0.35
    "minRelevanceScore": 70        // Only index patterns with score >= 70
  }
}

When to enable:

• Your queries use conceptual terms that don't match exact keywords
• You want more intelligent, context-aware search results
• You're okay with slightly slower first-time searches (embeddings need to compute)

Note: Requires downloading a ~50MB transformer model on first use. Embeddings are cached for performance.