dondai1234/master-fetch

What it solves

Hound provides AI agents with a reliable, cost-free way to access the live web. It eliminates the need for expensive third-party scraping APIs or complex account setups by providing a local Model Context Protocol (MCP) server that handles searching, fetching, and crawling while bypassing bot detection walls (like Cloudflare).

How it works

Hound operates as an MCP server that integrates with agents (such as Claude Code or Cursor). It uses a two-tier approach for web access: it first attempts a fast HTTP request and, if blocked, escalates to a "stealth" anti-detect browser (based on Patchright) that simulates human behavior to bypass bot walls. For search, it runs up to 10 keyless search engines in parallel on the user's machine, merging and ranking results using a local ONNX cross-encoder for semantic reranking.

Who it’s for

Developers building or using AI agents that require real-time web research, data extraction, and the ability to navigate websites that employ strict anti-bot measures without incurring per-request costs.

Highlights

• Keyless Local Search: Parallel search across 10 backends with neural reranking and cross-backend consensus for authority signals.
• Anti-Bot Stealth Engine: Bypasses Cloudflare Turnstile and other bot walls using human-like mouse movements and fingerprint rotation.
• MCP Integration: Designed as a single server that provides six specialized tools (smart_fetch, smart_crawl, smart_search, screenshot, cache_clear, version) to any MCP-compatible agent.
• BYOK Support: Allows users to add their own API keys (Serper, Tavily, Exa, etc.) for primary search, with automatic rotation and fallback to local keyless engines.
• Crawl & Fetch: Supports bulk fetching, PDF reading with OCR, and best-first same-domain crawling converted to markdown.