mcp-deep-research

What is mcp-deep-research?

mcp-deep-research is a lightweight Model Context Protocol (MCP) server that helps you discover, retrieve, and read academic papers from the web using the Google Scholar interface. It is designed for researchers and anyone who needs to quickly search for and extract clean-text content from academic papers.

How to use mcp-deep-research?

Install via pip (command not yet listed; use pip install placeholders as the README does not provide the exact package name). Optionally launch Google Chrome with remote debugging on port 9222 if it is not already running. Then run the server with either mcp-deep-research --transport sse --port 8000 for SSE or mcp-deep-research --transport stdio for STDIO transport. The server exposes three read-only tools to any MCP-capable client.

Key features of mcp-deep-research

• Search Google Scholar with optional year filter and date sort
• Convert any public web page to clean Markdown
• Auto-detect paper content (title, abstract, body, references) and strip noise
• 100% read-only; no writes, uploads, or local file access
• Supports configurable proxy for institutional or regional firewalls
• Respects each site’s robots.txt

Use cases of mcp-deep-research

• Quickly find recent academic papers on a specific topic by year and date
• Fetch and read the full text of a paper directly in a chat interface
• Extract clean Markdown content from any public web page for further processing

FAQ from mcp-deep-research

What are the prerequisites?

Python ≥ 3.10 and Google Chrome/Chromium are required. Set the environment variable CDP_ENDPOINT to the Chrome debugging port (default http://localhost:9222). Optionally set GOOGLE_PROXY for an HTTP(S) proxy.

How do I configure the server?

Set the CDP_ENDPOINT environment variable to your Chrome debugging endpoint. Optionally set GOOGLE_PROXY for proxy support. You can also manually launch Chrome with --remote-debugging-port=9222 if the auto-launch fails.

Is mcp-deep-research read-only?

Yes, it is 100% read-only. It does not write, upload, or access local files, and it respects each site’s robots.txt.

What transports does mcp-deep-research support?

It supports two transports: SSE (set with --transport sse --port 8000) and STDIO (--transport stdio).

Does mcp-deep-research store any data?

No, the README states it is 100% read-only; no writes, no uploads, no local file access. No data is stored by the server itself.