Mercury Parser

What is Mercury Parser?

Mercury Parser fetches web pages with a headless Chromium (via Playwright) to capture JavaScript-rendered content, then uses the Mercury Parser to extract the article and returns it as Markdown. It follows pagination links automatically. Designed for AI agents like Claude to read and summarize articles.

How to use Mercury Parser?

The server exposes a single tool called extract which takes a URL and optional proxy argument. It can be used via Claude Desktop (local stdio) by adding the configuration to claude_desktop_config.json, or run over Streamable HTTP for remote access. It can also be tested via CLI with node extractContent.js <url> [proxy].

Key features of Mercury Parser

• Automatically fetches web pages with headless Chromium for JS-rendered content
• Extracts articles as Markdown including title, author, date, summary, and content
• Follows pagination links automatically to capture full articles
• Falls back to direct fetch when Playwright fails, ensuring partial content
• Supports per-call proxy override and default proxy via environment variable
• Available over stdio or Streamable HTTP transport

Use cases of Mercury Parser

• An AI agent reading and summarizing a JavaScript-heavy news article
• Extracting clean article content from a multi-page blog post with pagination
• Archiving articles as Markdown for offline reference or translation
• Integrating a content extraction pipeline with Claude or other MCP clients

FAQ from Mercury Parser

What are the runtime requirements?

Node.js 18+ and Playwright Chromium (auto-installed on first launch if missing).

Does it support proxy for extraction?

Yes. A default proxy can be set via the MERCURY_PROXY environment variable, and each call can override it with the proxy tool argument.

What happens if the headless browser fails?

On Playwright failure (e.g., HTTP/2 protocol errors, navigation timeouts), it falls back to a direct fetch using Mercury Parser so partial content is still returned.

What transports are available?

The server supports both stdio (default for local use) and http (Streamable HTTP) by setting the MCP_TRANSPORT environment variable to http.

Where is the article data stored?

The server only extracts and returns the article content in the response; it does not persist data locally or on a server.