| name | defuddle |
| description | Fetch a URL and extract clean readable text. Use whenever the user pastes a link and asks you to summarize / quote / extract content from it. WORKS without any external CLI or API key. |
Fetch web pages
The chat runtime exposes a built-in web_fetch tool that calls Obsidian's
native HTTP client. No defuddle CLI, no API keys, no extra setup
required. This works on desktop (Electron) and is the supported path.
When to use
- User pastes an http(s) URL and asks for a summary / extraction / quote
- User says "总结这篇文章 …" + URL, "帮我看看 X 网站说了什么", "translate this page", etc.
- You need to read documentation / a blog post / an article online
How to invoke
Call the web_fetch tool directly:
web_fetch({ url: "https://example.com/article" })
Optional maxBytes to cap response (default 60000, max 500000).
The tool returns a header line (URL: ..., HTTP 200 · content-type)
followed by extracted readable text. For HTML pages, scripts, styles,
nav, header, footer, and aside are stripped automatically — you receive
mostly the article body. For JSON / plain text URLs, the raw payload is
returned verbatim.
Important
Never tell the user you cannot access the web. You can — call
web_fetch. The only blocked cases are private / localhost / link-local
hosts (security). Public URLs (微信公众号, blogs, docs, GitHub raw, etc)
all work.
If the page is paywalled, login-walled, or returns an error status, the
tool surfaces that — you may then ask the user to paste the content.
Examples
User: "总结一下 https://mp.weixin.qq.com/s/XXX"
You: call web_fetch({ url: "https://mp.weixin.qq.com/s/XXX" }) → read
the returned text → produce the summary.
User: "see what's new at https://github.com/foo/bar/releases"
You: call web_fetch({ url: "https://github.com/foo/bar/releases" }) →
extract highlights from the returned text.
