What it does

Markurl is a Node library that fetches a URL and returns a clean Markdown representation plus normalized metadata (title, byline, publish date, excerpts, images, links) suitable for LLM context. It trims output to a token budget by cutting at section boundaries, handles different content types (articles via Readability, GitHub READMEs via raw fetch, generic <p>-density fallback), and merges metadata from JSON-LD, OpenGraph, Twitter cards, and HTML meta tags. Tracking parameters are stripped from extracted links. The library ships with an MCP server, making it directly usable as a tool from Claude Desktop, Claude Code, or any MCP client.

Who it's for

Developers building AI agents, chatbots, or AI-powered applications that need to fetch and summarize web content. Specifically for agents that incorporate recent web information into Claude's context—e.g., newsfeeds, research summaries, real-time data lookups.

Common use cases

• Fetch a news article and return its Markdown + metadata within a 4000-token budget for Claude analysis.
• Extract GitHub repository READMEs as structured Markdown for documentation lookups.
• Pre-process search results before feeding them into an LLM prompt.
• Build an agent that can read web pages directly without requiring a separate browser or HTML parsing tool.
• Supplement Claude's knowledge with real-time web content while keeping token usage predictable.

Setup pitfalls

• Requires Node 20+. Older versions will fail silently during install.
• The optional Playwright fallback (useBrowser: "auto") requires installing the playwright peer dependency separately and running npx playwright install chromium—skipping this means JS-heavy sites will return incomplete content.
• Rate limiting is built-in (500 ms minimum between requests to the same host by default); if fetching many URLs from the same domain in parallel, expect staggered responses.
• The fetch cache has a 5-minute TTL and 200-entry limit; high-volume agents should implement a persistent cache backend (Redis, etc.) via the FetchCache interface.