kagi

This MCP server exposes a dedicated tool named kagi_web_search that uses the Kagi search API to perform web queries on behalf of an AI assistant. It supports multiple search queries in parallel and formats the results for easy consumption by MCP clients. The server is built with TypeScript and the official MCP SDK, and ships with pre-built JavaScript files so you can run it without building locally. To use it, start the server (for example via npx github:elliottlawson/kagi-mcp-server) with your KAGI_API_KEY set in the environment. Once running, you can issue MCP requests that trigger kagi_web_search to perform web searches and return structured results suitable for follow-up reasoning or answer generation. The tool is specifically named kagi_web_search to avoid conflicts with other search tools in your MCP environment.

Prerequisites:

• Node.js 18 or higher
• Access to the Kagi API (KAGI_API_KEY)

Installation and setup (two common paths):

Option A: Quick start with npx (no local install required)

1. Ensure you have your API key from Kagi.
2. Run the server directly: KAGI_API_KEY=your_api_key_here npx github:elliottlawson/kagi-mcp-server

Option B: Clone and run locally

1. Clone the repository and install dependencies: git clone https://github.com/elliottlawson/kagi-mcp-server.git cd kagi-mcp-server npm install
2. Run the pre-built server directly (no build step needed): KAGI_API_KEY=your_api_key_here node build/index.js
3. If you modify the TypeScript source, rebuild first: npm run build npm start

Notes:

• The repository includes pre-built JavaScript files, so you can run without a build step. If you clone and modify, you’ll typically run npm install and npm run build before starting the server.

Tips and notes:

• Set KAGI_API_KEY in your environment before starting the server.
• The server communicates via stdio with MCP clients.
• The tool name is kagi_web_search to avoid conflicts with other search tools.
• If you encounter issues, the MCP Inspector can be used to debug; run: npx @modelcontextprotocol/inspector npx github:elliottlawson/kagi-mcp-server or npx @modelcontextprotocol/inspector node build/index.js and then open http://localhost:5173 to inspect.
• Ensure your API access to Kagi is active (beta access may be required).
• When configuring in Claude or Cursor, you can pass the KAGI_API_KEY via env or config to ensure the server authenticates properly.