websearch MCP Server

Multi-engine web search MCP server for OpenCode. Searches DuckDuckGo, Google AI Mode (udm=50), Tavily, Wikipedia, arXiv, and AnySearch in parallel with embedding dedup + cross-encoder reranking.

Tools

• search — Multi-engine search with dedup + reranking (depth param controls snippet-only vs full fetch)
• fetch — URL to markdown/text via CDP-first (Helium browser) with Cloudflare bypass
• crawl — BFS/DFS deep crawl via Crawl4ai (CDP-connected Helium)
• screenshot — CDP screenshot or ARIA accessibility snapshot
• wikipedia — Search, summaries, geosearch, categories, pageviews, revisions
• arxiv — Academic paper search with full metadata
• map_site — URL structure discovery via sitemap XML (supports RSS/Atom/plain text/gzipped), robots.txt, and HTML link fallback
• pdf_extract — Full PDF extraction (tables, OCR, formulas) via opendataloader-pdf

Setup

1. Copy run.example to run and fill in API keys
2. Install dependencies in a Python 3.14+ venv
3. Run via OpenCode MCP config pointing to run