| name | extract |
| description | Extract useful information from URLs, including articles, docs, product pages, GitHub repositories, announcements, and reference pages. Use when the user asks to extract, vytěžit, summarize, process, or turn a URL into structured notes, facts, wiki input, or Markdown. |
| allowed-tools | Bash,Write,Read,WebFetch,WebSearch |
Extract
Extract information from a URL into clean, reusable Markdown. Prefer durable facts, technical details, key differentiators, examples, and source-backed claims over marketing copy.
Workflow
-
Fetch the URL with WebFetch when possible.
-
If the page is an article or blog post and cleaner Markdown is needed, use mreader:
npx mreader "URL"
-
Extract only information with lasting reference value:
- names, URLs, authors, dates, versions, platforms, licenses
- concrete capabilities, constraints, and integration points
- commands, APIs, options, examples, and important terminology
- source attribution when a claim is not already tied to the linked URL in the content
-
Discard launch hype, pricing/free-tier details, repeated navigation text, cookie banners, social widgets, and unsupported claims.
-
If the user wants wiki processing, hand the result to the /process workflow instead of saving a standalone extraction.
mreader options
-o file.md — save to file
-j — output JSON with title, author, content, and metadata
--no-frontmatter — omit YAML frontmatter
--no-content-negotiation — skip Accept: text/markdown, always extract from HTML
--url <url> — base URL for stdin mode (curl ... | mreader - --url <url>)
Output
Unless the user asks for a file, respond with:
## Summary
[2-4 sentences with the main point]
## Extracted facts
- [Fact with relevant source link if needed]
- [Fact with relevant source link if needed]
## Useful details
- [Commands, APIs, options, examples, or constraints]
## Notes
- [Unclear, missing, or time-sensitive information, if any]