Pure.md MCP Server: Markdown with Proxy Rotation
Fetch web pages as Markdown with Pure.md MCP server, avoid bot detection, use automatic proxy rotation and built-in headless JS rendering.
npx -y @puremd/puremd-mcpOverview
Pure.md MCP Server exposes pure.md’s web-fetching and markdown conversion capabilities to MCP-compatible clients (Model Context Protocol). It acts as a local MCP server that teaches clients how to fetch webpages, render JavaScript-heavy sites, and convert HTML/PDF/images into clean Markdown. The server includes automatic proxy rotation and other anti-bot measures so agents can reliably retrieve content that would otherwise be blocked.
Think of it as a CDN-like markdown delivery layer for LLMs: requests are proxied through pure.md, transformed into Markdown, and globally cached for later reuse. This reduces load on origin sites, helps avoid bot detection, and gives local tools consistent, readable web content for summarization, search, and reasoning workflows.
Features
- Headless JavaScript rendering for SPAs and dynamic pages
- Automatic proxy rotation and anti-bot techniques to reduce blocks
- Converts HTML, PDFs, and images into Markdown-friendly output
- Global caching (CDN-like) so repeated requests are fast and origin-friendly
- MCP-compatible server: plug into Cursor, Windsurf, Claude Desktop, and other MCP clients
- Optional API key for higher rate limits; anonymous usage supported
- Includes two CLI/helpers: unblock-url and search-web for common tasks
Installation / Configuration
Install and run the MCP server via npx (no global install required):
# Run once via npx
Set an API key (optional — leave empty to use anonymously). The server reads PUREMD_API_KEY from environment:
| Env var | Description |
|---|---|
| PUREMD_API_KEY | Optional API token from https://pure.md. Leave as “” to run anonymously. |
Example (bash):
Client configuration examples — add an MCP server entry that launches the local MCP implementation using the same npx command.
Cursor (~/.cursor/mcp.json):
Windsurf (./codeium/windsurf/model_config.json):
Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
Install via Smithery (automates client install for Claude):
Available Tools
The MCP package ships with helper utilities designed for common tasks. Invoke them via npx or through MCP clients that execute the binary.
- unblock-url — Fetch a single URL and return the page rendered and converted into Markdown while using proxy rotation to avoid blocks.
- search-web — Run a web search for a query and concatenate the most relevant results into a single Markdown document.
Example CLI usage:
# Convert a page to Markdown
# Perform a search and get combined markdown results
(Exact CLI flags may evolve; check the repository for the latest usage.)
Use Cases
- LLM context augmentation: Provide long-form, clean Markdown of web pages to an LLM prompt to improve retrieval augmentation and summarization.
- Agent web access: Equip local agents (Cursor, Claude Desktop, etc.) with an unblocker so they can fetch pages despite site bot protections.
- Research and note-taking: Convert academic articles or docs (including PDFs) to Markdown for easier annotation, diffs, and syncing with note apps.
- Web scraping for content extraction: Quickly produce Markdown-friendly versions of pages for downstream parsing, citation, or archiving.
- Search aggregation: Aggregate search results into a single markdown document for faster review or prompt consumption.
Resources
- Project on GitHub: https://github.com/puremd/puremd-mcp
- Pure.md (service and API): https://pure.md
- Model Context Protocol (MCP) overview: https://modelcontextprotocol.io/introduction
If you are new to MCP, start by installing via npx with an optional PUREMD_API_KEY, add the server entry to your MCP client’s configuration, and test with unblock-url on a known site. The repository contains the latest command and subcommand documentation.