Swift Online Search MCP

Swift Online Search MCP is a stdio MCP server for online search, latest facts, social checks, release research, and web research. It exposes a single OnlineSearch tool that delegates search orchestration to the internal Swift tool library.

Installation

Via npm

No build step is required when using the npm package. It ships prebuilt binaries for macOS Apple Silicon, Linux x64, and Linux arm64.

npx @mehmetbaykar/swift-online-search-mcp

Or install it globally:

npm install -g @mehmetbaykar/swift-online-search-mcp
swift-online-search-mcp

Build From Source

Requires macOS 14+ and Swift 6.2+.

git clone https://github.com/mehmetbaykar/swift-online-search-mcp.git
cd swift-online-search-mcp
swift build
.build/debug/swift-online-search-mcp

Configuration

The server requires provider credentials before startup:

• PROXY_BASE_URL
• PROXY_API_KEY
• POE_BASE_URL
• POE_API_KEY

Production reads configuration from ~/.super-code/.env, with process environment variables taking precedence. Local development can use .env.local.

MCP Client Setup

Claude Code:

claude mcp add swift-online-search -- npx -y @mehmetbaykar/swift-online-search-mcp

Codex:

codex mcp add swift-online-search -- npx -y @mehmetbaykar/swift-online-search-mcp

Manual JSON configuration:

{
  "mcpServers": {
    "swift-online-search": {
      "command": "npx",
      "args": ["-y", "@mehmetbaykar/swift-online-search-mcp"]
    }
  }
}

Tools

OnlineSearch

OnlineSearch accepts a required query and an optional route-hint mode:

• automatic (default): GPT-5.6 Sol chooses the smallest useful route.
• fast: starts with Perplexity Sonar.
• technical: starts with Perplexity Sonnet Thinking.
• latest: combines Perplexity Sonar with Grok web/X coverage when useful.
• deep: combines Perplexity Deep Research with Grok.

Modes are hints, so one request can combine routes. Ordinary research uses Perplexity GPT Thinking. Specialized LinkedIn, URL extraction, similarity, company, code, Exa, Linkup, Reka, or Poe tools are exposed only when the query indicates that need.

GPT-5.6 Sol is the sole orchestrator, using medium reasoning and the Responses API. A failed delegated tool is removed by its exact tool ID and the request is replanned once. Response budgets range from 4,000 tokens for fast search to 16,000 for deep research; delegated results are truncated to a mode-specific context budget before synthesis.