The server is open source under the MIT license, but it calls Kagi's paid Search API, so you need a Kagi account and a KAGI_API_KEY with available credits. There is no free query tier through this server.

Does it support remote/OAuth?

By default it runs locally over stdio via uvx and authenticates with an API key rather than OAuth. It can also run in HTTP mode with the --http flag for multi-tenant or self-hosted deployments, with a Docker image provided.

What happened to the FastGPT and Summarizer tools?

The earlier kagi_fastgpt and kagi_summarizer tools have been removed from the current server and are planned to return in a future release. Today the server ships kagi_search_fetch and kagi_extract.

Kagi MCP server

OfficialKagi402Config last verified Jun 1, 2026

Kagi's official MCP server gives agents ad-free, privacy-respecting web search plus clean full-page extraction.

Kagi MCP is the official Model Context Protocol server from Kagi, the paid, ad-free search engine built around privacy and result quality. It lets an AI agent run searches against Kagi's index and pull back clean, ready-to-reason-over content rather than a page of ads and tracker-laden links, which makes it a strong grounding tool for research, fact-checking, and answering questions about current events beyond a model's training cutoff. Because Kagi is subscription-funded, results are not shaped by advertising, and the same account that powers your browser search powers the agent.

The server exposes two tools: kagi_search_fetch runs web, news, video, podcast, and image search with optional page extracts, domain and date filters, and Kagi lenses, while kagi_extract fetches a single known URL's full content as markdown. It is published as the kagimcp Python package and runs locally over stdio with uvx, authenticating with a KAGI_API_KEY from your Kagi account. It can also run in HTTP mode via a --http flag for multi-tenant or self-hosted deployments, with a Docker image provided for platforms like Railway, Render, Cloud Run, and Fly.io.

Quick install

Copy-paste configs are provided for all 8 supported clients. Pick your client below.

Add to ~/.claude.json

~/.claude.json

json

{
  "mcpServers": {
    "kagi": {
      "command": "uvx",
      "args": [
        "kagimcp"
      ],
      "env": {
        "KAGI_API_KEY": "<KAGI_API_KEY>"
      }
    }
  }
}

Or via CLI

bash

claude mcp add kagi -- uvx kagimcp

Step-by-step guides

Available tools

Tool	Description
kagi_search_fetch	Runs web, news, video, podcast, and image search with optional page extracts, domain and date filters, and Kagi lenses.
kagi_extract	Fetches the full content of a known URL and returns it as clean markdown.

Required configuration

KAGI_API_KEYRequired
Kagi API key used to authenticate search and extract requests.

What you can do with it

Ad-free research grounding

The agent searches Kagi and gets clean, ad-free content back, so it can synthesize an answer from quality sources without a separate scraping step or wading through SEO spam.

Read a specific page

Point kagi_extract at a known URL to pull the full article as markdown, giving the agent exactly the source text it needs to quote or summarize.

FAQ

Is it free?: The server is open source under the MIT license, but it calls Kagi's paid Search API, so you need a Kagi account and a KAGI_API_KEY with available credits. There is no free query tier through this server.
Does it support remote/OAuth?: By default it runs locally over stdio via uvx and authenticates with an API key rather than OAuth. It can also run in HTTP mode with the --http flag for multi-tenant or self-hosted deployments, with a Docker image provided.
What happened to the FastGPT and Summarizer tools?: The earlier kagi_fastgpt and kagi_summarizer tools have been removed from the current server and are planned to return in a future release. Today the server ships kagi_search_fetch and kagi_extract.

← Browse all search-and-data servers