Jina AI Reader

This is a lightweight MCP server that fetches and serves Markdown versions of web pages using r.jina.ai. It enables LLMs to access clean, token-aware web content with minimal overhead.

Could also use the existing fetch mcp, but where's the fun in that!?

Features

• Fetch any URL as Markdown
• Auto-hashes, tokenizes, and caches content
• Assigns stable URIs for resource lookup

Usage

To test locally:

npx @modelcontextprotocol/inspector uv run src/jina-mcp/server.py

Integrate with an mcp client for LLM use. e.g. claude, cursor, custom.

For example your config might look like this

{
  "mcpServers": {
    "jina-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "C:\\Users\\jswords\\mcp-jina\\src\\jina-mcp\\",
        "run",
        "server.py"
      ]
    }
  }
}

To Do

• Add support for batch fetching in parallel
• Deploy