# ac3xx/mcp-servers-kagi [Health: Active]

**Category:** 🔎 Search & Data Extraction  
**Repository:** https://github.com/ac3xx/mcp-servers-kagi  
**GitHub Stars:** 45  
**npm Downloads (last month):** 30651  
**Views:** 4  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/ac3xx-mcp-servers-kagi

## Description
A Model Context Protocol server implementation for Kagi's API

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "mcp-servers-kagi": {
    "command": "npx",
    "args": ["-y","@smithery/cli"],
    "env": {
      "KAGI_API_KEY": ""
    }
  }
}
```

**Requires environment variables:** `KAGI_API_KEY` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What ac3xx/mcp-servers-kagi MCP server does

ac3xx/mcp-servers-kagi MCP server exposes Kagi web search to an MCP-compatible client. The currently available capability is the `kagi_search` tool. It accepts a search query and an optional limit, sends the request to Kagi's Search API, and returns the resulting search data to the client.

This makes the project suitable for agent workflows that need to retrieve current web information before summarizing, comparing, or analyzing it. The server itself does not provide the analysis step; the connected MCP client can use the returned results as context.

The README identifies Kagi's API as being in private beta. Access therefore depends on having a Kagi API key and access to that API.

## How it works

The project is implemented in TypeScript and communicates with MCP clients over standard input and output. After the server is built and connected, a client such as Claude Desktop can invoke `kagi_search` during a conversation. The tool request contains the search text and may include a result limit. The server uses the configured Kagi credential to query the API and sends the response back through MCP.

The repository also names `kagi_summarize`, `kagi_fastgpt`, and `kagi_enrich` as future tools. They are not implemented, so their names should not be included in a working client configuration or workflow.

## Setup and configuration

Clone or otherwise obtain the repository, install its dependencies with `npm install`, and build it with `npm run build`. The generated server is then configured as an MCP process. The README also provides `npm run watch` for development rebuilds.

Set `KAGI_API_KEY` in a `.env` file at the project root, or pass the variable through the MCP client's environment configuration. Keep the file out of version control; the README specifically recommends adding `.env` to `.gitignore`.

For Claude Desktop, the documented configuration uses the built `build/index.js` entry point and supplies `KAGI_API_KEY` under the server's `env` object. The exact executable path must be replaced with the path on the host running ac3xx/mcp-servers-kagi MCP server. The repository also documents Smithery as an installation route for Claude Desktop, but it does not provide a published package command for direct installation of this repository.

## Tools and capabilities

- `kagi_search`: performs a web search through Kagi and accepts a query plus an optional limit.
- Returns search results that a connected agent can summarize or analyze.
- Supports local development with automatic rebuilds through the repository's watch script.
- Includes an Inspector script for debugging the server's stdio communication.

## Limitations and notes

Only web search is implemented. The planned summarization, FastGPT, and enriched-news operations cannot currently be called. The README also notes that error handling and input validation are areas for future improvement, so consumers should account for the project's development status.

The ac3xx/mcp-servers-kagi MCP server requires a Kagi API key and access to Kagi's private-beta API. It is not described as a hosted endpoint; the documented setup builds and runs the TypeScript project locally. Debugging can be performed with `npm run inspector`, which exposes browser-based inspection tools for the stdio connection.

_Full upstream README: https://allmcps.com/mcp/ac3xx-mcp-servers-kagi/readme_