Docs Fetch MCP Server

What is Docs Fetch MCP Server?

An MCP server that enables LLMs to fetch web content with recursive exploration. It retrieves clean, readable text from web pages and autonomously follows same-domain links up to a specified depth, making it ideal for learning topics by exploring documentation.

How to use Docs Fetch MCP Server?

Install via npm and build the project. Configure your MCP client (e.g., Claude Desktop) with the node command pointing to the built index.js. The server exposes a single tool fetch_doc_content which accepts a URL and optional depth (1–5), returning page content and discovered links.

Key features of Docs Fetch MCP Server

• Clean content extraction, removing navigation and ads
• Recursive same-domain link exploration (depth 1–5)
• Dual‑strategy: fast axios first, puppeteer fallback
• Global timeout handling to stay within MCP limits
• Parallel crawling with robust error handling

Use cases of Docs Fetch MCP Server

• An LLM learning a new framework by exploring its official docs
• Gathering comprehensive information from a documentation site
• Automatically collecting related pages from a single starting URL
• Enabling AI assistants to research topics with minimal user guidance

FAQ from Docs Fetch MCP Server

What is the main purpose of this server?

It lets LLMs autonomously fetch and explore web content, especially documentation, by recursively following same‑domain links to gather complete context on a topic.

What dependencies does it require?

Node.js, npm, and the packages @modelcontextprotocol/sdk, puppeteer, and axios. Puppeteer is used as a fallback for complex pages.

How do I install and configure it?

Clone the repo, run npm install and npm run build, then add the server to your MCP client’s configuration with the node command pointing to build/index.js and set MCP_TRANSPORT to "pipe".

What recursion depth can I use?

The depth parameter accepts 1 to 5, with a default of 1. Higher values explore more linked pages within the same domain.

Does the server handle errors or timeouts?

Yes. It implements global timeout handling, graceful error recovery for network issues, and returns partial results even if some pages fail.