jina-ai/MCP

What it solves

This project provides a remote Model Context Protocol (MCP) server that allows AI agents and LLM clients (like Claude Desktop, Cursor, and LM Studio) to access Jina AI's suite of search, reading, and document processing tools. It bridges the gap between an LLM's internal knowledge and real-time web data, academic papers, and complex document layouts.

How it works

The server implements the MCP standard using Streamable HTTP transport. It exposes a set of tools that the LLM can call to interact with Jina AI's APIs. Users can configure their clients to connect to the remote server at https://mcp.jina.ai/v1 or deploy their own instance via Cloudflare Workers.

To manage context window limits, the server supports server-side filtering via query parameters (e.g., include_tools or exclude_tools), allowing users to only register the tools they actually need.

Who it’s for

Developers and AI agent users who want to give their LLMs the ability to search the web, read academic papers from arXiv and SSRN, extract content from PDFs, and rerank documents for relevance.

Highlights

• Comprehensive Toolset: Includes tools for web search, arXiv/SSRN search, reading URLs as markdown, capturing screenshots, and extracting PDF figures and tables.
• OCR Integration: Supports OCR for scanned documents and image-only PDFs to preserve formulas and table structures.
• Context-Aware Reading: The read_url tool can take a question parameter to return only the most relevant passages of a page rather than the full body.
• Flexible Configuration: Supports both direct remote connection and local proxying via mcp-remote for clients that do not support remote MCP servers.
• Token Management: Implements a token guardrail to prevent responses from being rejected by clients with fixed token limits (e.g., 25k tokens).