Hound MCP Server: Web Search and Page Fetching for Claude, Cursor and Other Apps

• Searches the web and returns links with snippets

  Runs a web search across many search engines at once and merges the results into one ranked list. It returns links and short snippets, not page text, so the assistant follows up by fetching the best ones. Search keys you add yourself are used first if you set them up.

  Try: “Search the web for recent benchmarks of local embedding models and give me the five best links.”

  Network · mcp_smart_search

• Reads a web page or PDF and returns clean text

  Loads a web address and returns its text as markdown. It starts with a plain request and switches to a stealth browser when the site blocks it. It can read PDFs by page, fetch several addresses in one call, keep only the passages that match a question, and click or fill in a page before reading it.

  Try: “Read https://example.com/whitepaper.pdf and pull out only the paragraphs about pricing.”

  Network · mcp_smart_fetch

• Walks through a site and reads many pages

  Starts at one address and follows links on the same site, returning each page as text. It can list a site's pages from its sitemap first so the assistant picks which ones to read. It stops at 10 pages and 2 levels deep unless told otherwise.

  Try: “Crawl the docs site at docs.example.com and read the pages about authentication.”

  Network · mcp_smart_crawl

• Takes a picture of a web page

  Opens a page in the stealth browser and returns a screenshot as an image. It suits pages where the layout, charts or images carry the meaning, and only works with assistants that can look at images.

  Try: “Take a screenshot of example.com/pricing so I can see how the plans are laid out.”

  Network · mcp_screenshot

• Deletes Hound's saved copies of pages

  Hound keeps fetched text on your computer for an hour so repeat requests are fast. This tool removes the expired copies, or every saved copy when told to wipe all of them. It does not touch anything outside Hound's own cache.

  Try: “Clear Hound's page cache so the next fetch is fresh.”

  Delete · cache_clear

• Shows the installed Hound version

  Reports which Hound version is running and whether a newer one exists on PyPI. Useful for checking the setup took effect.

  Try: “Which version of Hound is running, and is there an update?”

  Network · version

Watch out for

• Published from a different GitHub account

  The pip package links to dondai44423/master-fetch from version 13.0.0 on, while this page's main repo is dondai1234/master-fetch (897 stars). The changelog says the original account was locked and development moved. Verify before installing; our advice is to pin 12.4.1 if you want code that matches the original repo. github.com

• It can click and fill forms on web pages

  The fetch tool accepts click, fill, press, wait and scroll steps that run in a real browser before reading. The browser uses a throwaway profile, not your own, but a page you open can still receive form input. github.com

• The author says Hound is abandoned

  The fork README says Hound is no longer maintained and points to a Rust rewrite called DonSeTch. The last release was 13.2.0 on August 26, 2026, so expect no security fixes. github.com

• Fetched pages can carry hidden instructions

  Our advice: treat page text as untrusted, because a hostile page can include wording aimed at your assistant. Do not let it act on instructions it finds in fetched pages. github.com

• It ignores robots.txt unless asked

  Hound checks a site's robots.txt only when the request sets respect_robots. It also tries to get past bot checks, so follow each site's terms yourself. github.com

• Your search terms go to many search engines

  One search sends the query to eight search engines at once, plus Semantic Scholar, GitHub or Hacker News when the question fits. Do not search for anything you would not type into those sites. github.com

• It keeps a local cache and can delete it

  Fetched text is saved on your computer, and cache_clear deletes those saved copies. It does not delete anything outside Hound's own cache. github.com

• Search keys and proxies are stored as plain text

  If you add your own search keys or proxies, they are saved as plain JSON files in ~/.hound. Use file permissions or environment variables on a shared machine. github.com

Security advisories