dsh-webfetch

In the DeepSeek Harness ecosystem, agents excel at searching for information, but given a URL, they previously lacked the ability to read the webpage content behind it, creating a gap in comprehensive information retrieval. The dsh-webfetch plugin solves this by equipping agents with robust web reading capabilities, bridging the divide between search and deep understanding.

At its core, dsh-webfetch offers four read-only tools: web_fetch fetches a webpage and extracts clean Markdown or plain text, preserving headings, links, and code blocks for easy processing; web_links systematically lists all links on a page, aiding in site structure analysis; web_feed parses RSS 2.0 and Atom feeds into structured entries, handling XML complexities and supporting content extraction; and web_headers inspects HTTP status, response headers, and redirect chains without downloading the body, serving as a handy diagnostic tool. All tools adhere to strict safety principles: only http/https protocols are allowed, no credentials are ever sent, and built-in limits on timeouts, redirects, and content size ensure secure and controlled operations.

This plugin is ideal for anyone looking to enhance DeepSeek Harness agents' web interaction capabilities. Whether you're a developer building applications involving web analysis, content scraping, or RSS subscriptions, or a user wanting agents to handle online information more intelligently, dsh-webfetch integrates seamlessly with zero runtime dependencies, making it ready to use immediately after installation. It perfectly complements the built-in search functionality—search provides leads, while webfetch delivers deep reading, collectively boosting the agent's practicality and flexibility.