Jina

You can run a remote MCP server that exposes Jina Reader, Embeddings, and Reranker capabilities via a simple HTTP endpoint or through a local proxy. This lets your client apps query web content, search the web, and retrieve structured results through a unified set of tools.

How to use

Connect your client to the remote MCP endpoint to access Jina Reader, Embeddings, and Reranker tools. Use the HTTP connection for a direct URL-based MCP server, or use the local proxy approach if your client does not support remote MCP servers yet.

How to install

Prerequisites: ensure you have Node.js and npm installed on your machine. You can verify by running node -v and npm -v in your terminal.

Install and start the MCP server locally using the provided development workflow. The following commands will clone the MCP project, install dependencies, and start the development server.

# Clone the repository
git clone https://github.com/jina-ai/MCP.git
cd MCP

# Install dependencies
npm install

# Start development server
npm run start

Configuration and usage notes

Two ways to connect to the MCP server are documented here. You can use a remote HTTP endpoint or a local stdio proxy to reach the same remote server.

# HTTP (remote MCP server)
{
  "mcpServers": {
    "jina_mcp": {
      "url": "https://mcp.jina.ai/sse",
      "headers": {
        "Authorization": "Bearer ${JINA_API_KEY}"
      },
      "args": []
    }
  }
}

Security and API keys

If you are using a remote MCP server, you can supply an API key for higher rate limits and better performance. Use a bearer token in the Authorization header when connecting to the HTTP endpoint.

Troubleshooting

If you encounter a tool calling loop or unexpected behavior, ensure your model maintains sufficient context length for the tool chain. For LMStudio workflows, consider adjusting the model choice or context window to prevent loss of track during long tool sequences.

Notes on tooling availability

A suite of tools is exposed by the MCP server, including URL-to-markdown extraction, web and arXiv search, image search, query expansion, and deduplication utilities. These tools can be mixed and matched to build complex workflows across content extraction, search, and ranking tasks.

Developer guidance

If you want to run the MCP server in a cloud-friendly environment, you can deploy the server to a CDN or edge compute platform that supports HTTP endpoints for MCP. When running locally, the standard Node.js tooling (npm) is used to install, build, and start the server.

Available tools

primer

Get current contextual information for localized, time-aware responses.

read_url

Extract clean, structured content from web pages as markdown via the Reader API.

capture_screenshot_url

Capture high-quality screenshots of web pages via the Reader API.

guess_datetime_url

Analyze web pages for last update/publish datetime with confidence scores.

search_web

Search the entire web for current information and news via the Reader API.

search_arxiv

Search academic papers and preprints on arXiv via the Reader API.

search_images

Search for images across the web via the Reader API.

expand_query

Expand and rewrite search queries with the Reader API's expansion model.

parallel_read_url

Read multiple web pages in parallel for efficient content extraction.

parallel_search_web

Run multiple web searches in parallel for comprehensive topic coverage.

parallel_search_arxiv

Run multiple arXiv searches in parallel for diverse academic angles.

sort_by_relevance

Rerank documents by relevance to a query via the Reranker API.

deduplicate_strings

Get top-k semantically unique strings via Embeddings API and submodular optimization.

deduplicate_images

Get top-k semantically unique images via Embeddings API and submodular optimization.