mdget.dev
Open source · Apache-2.0 · 16 platform adapters

Pages your agent can't open

Put this host in front of the link. Ordinary pages, and the ones that keep a reader out — WeChat, Zhihu, Feishu — come back as Markdown. For you, and for an agent.

GET https://mdget.dev/https://example.com

Optional: format, selector, force_browser, raw, engine

Try a Cloudflare docs pageSee pages that block youAPI reference

16 Platform Adapters

WeChatZhihuFeishuYuqueJuejinCSDN36KrToutiaoWeiboNetEaseTwitter/XRedditNotionGitHubSubstackMedium

Proof

The link is blocked. The article comes back.

A WeChat article, as people usually meet it. Left is the page. Right is the Markdown from the same conversion.

What the page shows

WeChat Official

Deep Dive: LLMs in Enterprise Production

Lin Ke, 2026-03-25

This content requires the WeChat app

Scan QR code

Markdown you get
# Deep Dive: LLMs in Enterprise Production

> Lin Ke, 2026-03-25

## Key Takeaways

1. LLMs are moving from lab experiments to production environments
2. RAG architecture is the preferred approach for enterprise apps
3. Agent workflows will redefine software development

## Background

With the release of GPT-4, Claude, and other models, enterprises are seriously considering integrating LLMs into their core business workflows...

SPAs, paywalls, anti-bot pages

Sixteen adapters cover WeChat, Zhihu, Feishu, Twitter and more. When a static fetch comes back empty, it escalates to a headless browser.

Reachable from your agent

An MCP server, Agent Skills, and llms.txt ship with it. Point Claude, Cursor, or your own agent at the host — no glue code.

Five fallbacks, in order

Native extraction, Readability, a headless browser, Firecrawl, then Jina. Browser rendering and external readers require an API key. Results are cached at the edge.

How it works

One prefix. The edge does the rest.

  1. 1

    Prepend URL

    Add mdget.dev/ before any web address

  2. 2

    Edge Pipeline

    Tried in order: native extraction, Readability, a browser, Firecrawl, then Jina. Browser rendering and external readers require an API key.

  3. 3

    Clean Output

    Markdown, JSON, HTML, or plain text

Five fallback layers

Each layer is tried in order. The first one that returns content wins.

01Native MarkdownCloudflare edge, about 0.1s
02ReadabilityHTML parsing, about 0.5s
03Headless browserabout 2–5s
04Cloudflare RESTabout 1–3s
05Firecrawl, then JinaExternal fallback, about 2–4s

What people use it for

AI Agents

Feed web content to LLMs in clean Markdown

Knowledge Base

Crawl docs, wikis, blogs with deep crawl

Content Migration

Batch convert up to 10 URLs

Research

Read any article, no login walls

Chinese Web

WeChat, Zhihu, Feishu, Yuque, CSDN...

Data Extraction

CSS selectors, XPath, or Regex

Pick up where you are

You have a link

Paste it above. Markdown opens on the next page.

An agent should read the web

Skills, MCP, or point it at llms.txt.

You are wiring a pipeline

Batch, extract, jobs, and deep crawl are in the docs. Keys live in the portal.

Frequently asked questions

Paste a URL.

A free key unlocks browser rendering and higher limits.

Get API KeyRead the docs