A Model Context Protocol (MCP) server for integrating with Jina Reader API, allowing Claude and other AI assistants to read and extract content from webpages, HTML, and PDF files.
- Read content from URLs
- Process HTML content directly
- Extract information from PDF files (base64 encoded)
- Format responses with links and metadata
npm install -g @kennyfrc/jina-reader- Clone this repository
- Install dependencies:
npm install
- Build the project:
npm run build
The easiest way to use this package is with npx. You need to provide your Jina API key as an environment variable:
# Set the API key directly in the command
JINA_API_KEY=your_jina_api_key npx -y @kennyfrc/jina-readerIf you've cloned the repository:
# Provide the API key when running
JINA_API_KEY=your_jina_api_key npm startAdd this to your claude-desktop-config.json:
{
"mcpServers": {
"jina-reader": {
"command": "npx",
"args": ["-y", "@kennyfrc/jina-reader"],
"env": {
"JINA_API_KEY": "your_jina_api_key_here"
}
}
}
}Or if you've cloned the repository:
{
"mcpServers": {
"jina-reader": {
"command": "node",
"args": ["/absolute/path/to/your/dist/index.js"],
"env": {
"JINA_API_KEY": "your_jina_api_key_here"
}
}
}
}Place this file in:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
Restart Claude Desktop to load your MCP server.
Each tool accepts the following optional parameters:
-
engine: Control the rendering engine:none: Default rendering engine (good balance of quality and speed)direct: Speed-optimized renderingbrowser: Quality-optimized renderingcf-browser-rendering: Experimental rendering
-
max_length: Maximum number of characters to return (default is 20000)- Important: This parameter is essential for controlling token usage, especially with large documents like research papers
- By default, content is truncated to 20,000 characters to prevent excessive token consumption
- The response includes pagination information when content is clipped
- Pagination details show the current character range and guidance for subsequent requests
-
start_index: Start content from the character index (default is 0)- Essential for paginating through large documents
- Works with the pagination info provided in clipped responses
- For example, to get the next page of content after seeing "Currently showing characters 0-20000", use
start_index=20000
-
jina_read_url: Read and extract content from a webpage{ "url": "https://example.com", "engine": "browser", // Optional, for best quality "max_length": 10000, // Optional, limit output length "start_index": 0 // Optional, pagination starting point }
-
jina_read_html: Read and extract content from HTML{ "html": "<html>...</html>", "engine": "direct", // Optional, for faster processing "max_length": 10000, // Optional, limit output length "start_index": 0 // Optional, pagination starting point }
-
jina_read_pdf: Read and extract content from a PDF file{ "pdf": "base64_encoded_pdf_content", "engine": "none", // Optional, uses default engine "max_length": 10000, // Optional, limit output length "start_index": 0 // Optional, pagination starting point }
When dealing with large documents (like research papers or lengthy articles), the content will be automatically clipped to control token usage. The response includes clear pagination information:
[PAGINATION INFO]
- Content clipped: Currently showing characters 0-20000 of approximately 78000 total
- To view the next section, use start_index=20000 with the same max_length
- Complete content can be accessed by making multiple paginated requests
To retrieve subsequent pages, make additional requests with updated start_index values:
// First request (page 1)
{
"url": "https://arxiv.org/html/2401.14196v1",
"max_length": 20000
// start_index defaults to 0
}
// Second request (page 2)
{
"url": "https://arxiv.org/html/2401.14196v1",
"max_length": 20000,
"start_index": 20000
}
// Third request (page 3)
{
"url": "https://arxiv.org/html/2401.14196v1",
"max_length": 20000,
"start_index": 40000
}This MCP server implements two levels of caching:
- Local Memory Cache: Responses are cached in memory for 3 hours to improve performance and reduce API calls
- Optimized for documentation content which changes infrequently
- Significantly reduces API usage for repeated queries
- Jina API Server Cache: The Jina API also caches responses using the
X-Cacheheader
The caching system helps to:
- Improve response times
- Reduce API usage
- Decrease latency for repeated requests
- Minimize costs when working with large documentation
For development with automatic rebuilding:
npm run devFor publishing instructions, see the docs/publishing.md file.