JinaAI

You can access Jina.ai’s web search via an MCP server that formats results for large language models. This server delivers fast, clean, and LLN-friendly web content, making it easier to retrieve relevant information from the web and present it in a structured way to your applications.

How to use

You run the MCP server locally or in your MCP environment and connect to it using an MCP client. Start the server in one of your preferred methods, then issue a search query through the MCP client. The tool returns the top results with URLs and clean content suitable for LLms. You can customize behavior with options like token budgeting, content gathering, and locale.

How to install

Prerequisites: you need Node.js installed on your system. Optionally, you can use WSL on Windows if you prefer to run commands through a Linux environment.

# Prerequisites check (example for Ubuntu)
node -v
npm -v

# Install dependencies and prepare the MCP environment (example steps)
npm install
pnpm install

# Start the server locally using Node
node -y mcp-jinaai-search

# Start the server in development mode (if you plan to develop or test locally)
pnpm run dev

Configuration and usage notes

Configure the MCP client to point to the Jina.ai search MCP server using one of the supported runtimes shown below. Your configuration will include the command to run the server and any required environment variables.

Additional configuration options

The MCP server exposes a single tool named search with configurable parameters. You can adjust query, response format, caching, token budget, locale, and content gathering behaviors to tailor results for your use case.

Security and keys

You must provide an API key for Jina.ai. Do not expose your API key in client configurations or logs. Store it securely and reference it in environment variables during server startup.

Troubleshooting

If the server fails to start, verify Node.js is installed, confirm the API key is accessible to the process, and check that the command arguments are correct for your runtime (local Node run vs WSL-based run). Review logs for any authentication or network errors.

Notes

If you prefer not to run the server locally, you can use a connected environment or a managed MCP runtime that supports stdio configurations. Ensure your environment variables are set in the respective session or startup script.

Available tools

search

Queries the web via Jina.ai to return the top results with URLs and clean, LLMe-friendly content. Supports options like format, cache control, token budget, locale, streaming, and gathering links/images.