master-fetch

What it is

Hound is a single MCP server that provides fetch, crawl, and local search capabilities with Cloudflare bypass, Trafilatura extraction, and a stealth browser. It runs locally without keys and is self-hosted under the MIT license.

How it works

• The tool performs HTTP fetches first; if blocked, it escalates to a Patchright anti-detect browser using a stealth engine with Cloudflare Turnstile solving. It supports both plain HTTP fetches and browser-based extraction via a single browser instance.
• Extraction is powered by Trafilatura (plus Markdownify) for content extraction during fetches and crawl results.
• It maintains a local SQLite cache (WAL mode) keyed by URL, extraction type, css_selector, and pages, with a size cap and eviction policy.
• For search, it runs multiple keyless backends in parallel, then merges and ranks results using a neural rerank model; results include relevance signals and per-result metadata.
• The crawler (smart_crawl) uses a best-first, same-domain strategy with content-adaptive extraction rules and optional sitemap mode.

Getting started

pip install hound-mcp[all]          # fetch + crawl + keyless search + PDF + OCR + neural rerank
playwright install chromium         # the anti-detect browser engine

Then point any MCP client at the hound command. No arguments, no keys, no env vars.

hound -v          # version + update status
hound -u          # update to latest (brick-proof, self-healing)
hound --doctor    # health check + fix advice
hound --rollback  # undo the last update

If issues arise, recover with python ~/.hound/repair.py or hound --doctor.

Getting started (continued)

• The project provides six tools: smart_fetch, smart_crawl, smart_search, screenshot, cache_clear, and version.
• Local search uses 10 backends in parallel with neural reranking and a cross-backend consensus. It returns URLs with ranking and relevance signals rather than full pages.

Getting started (installation notes)

• The Quick Start section shows the recommended installation commands and how to run the tools locally without keys.

Recent releases

• v11.1.6: Fix smart_crawl Cloudflare bypass; ensures Cloudflare Turnstile handling and internal extraction consistency.
• v11.1.5: Adjusted Response.css() to propagate errors instead of returning []. Removed broad except blocks that swallowed errors.
• v11.1.0: Next-gen stealth engine with memory optimization; stealth fetch and memory improvements.
• v11.0.3: Self-healing entry point and stale process cleanup.
• v11.0.2: Fixes to smart_fetch and smart_crawl after previous scrapping removal changes.

Traction

• Stars: 448
• Stars in last day: 224

Behind the repo

• The project is MIT-licensed and repository owner is dondai1234.

Caveats

• License: MIT
• Created: 2026-06-01
• Last push: 2026-07-21
• Open issues: 0
• Language: Python
• Requires Chromium via Playwright for the anti-detect browser.