Kagi MCP Server

Content

Kagi MCP Server connects Claude and other MCP clients to Kagi's Search and Extract APIs. It exposes kagi_search_fetch for web, news, video, podcast, and image searches with optional filters, date limits, domains, file types, lenses, and inline page extracts. It also exposes kagi_extract for retrieving a public page as markdown.

Use it when a Claude workflow needs current web results or page extraction from a Kagi-backed source instead of a generic scraping stack. For local desktop usage, run it over stdio with KAGI_API_KEY; for shared deployments, review the HTTP bearer-token flow and logging behavior before exposing the endpoint.

Source Review

• https://github.com/kagisearch/kagimcp
• https://raw.githubusercontent.com/kagisearch/kagimcp/main/README.md
• https://pypi.org/pypi/kagimcp/json
• https://raw.githubusercontent.com/kagisearch/kagimcp/main/LICENSE
• https://raw.githubusercontent.com/kagisearch/kagimcp/main/pyproject.toml
• https://raw.githubusercontent.com/kagisearch/kagimcp/main/src/kagimcp/server.py
• https://raw.githubusercontent.com/kagisearch/kagimcp/main/Dockerfile

These sources were reviewed on 2026-06-06. Prefer the live repository, README, PyPI metadata, license, package metadata, server implementation, and Dockerfile for current setup and operating details.

Features

• Search the web through Kagi's general search workflow.
• Search news, videos, podcasts, and images with workflow-specific result types.
• Fetch inline markdown extracts for top search results.
• Extract a specific HTTPS page as markdown through the Kagi Extract API.
• Restrict searches by included or excluded domains.
• Filter by relative time, explicit date bounds, file type, or Kagi lens.
• Hide selected optional search parameters from the LLM-facing schema with KAGI_HIDDEN_PARAMS.
• Configure API retries, search timeout, extract timeout, and log level.
• Run over local stdio for desktop clients or streamable HTTP for hosted use.
• Use bearer tokens per request in HTTP mode so one hosted instance can serve multiple Kagi users without a shared server-wide key.

Installation

Install uv, then add Kagi MCP to an MCP client:

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

Claude Code can add the same server from the command line:

claude mcp add kagi -e KAGI_API_KEY="YOUR_KAGI_API_KEY" -- uvx kagimcp

For local development from a checkout:

KAGI_API_KEY=YOUR_KAGI_API_KEY uv run kagimcp

For streamable HTTP mode:

KAGI_API_KEY=YOUR_KAGI_API_KEY uv run kagimcp --http --host 0.0.0.0 --port 8000

Use Cases

• Give Claude current web results without leaving the MCP client workflow.
• Search recent news or technical documentation with explicit domain and date filters.
• Fetch markdown extracts from public pages before summarizing or comparing sources.
• Use Kagi lenses to constrain research to curated source sets.
• Add web, news, video, podcast, or image discovery to coding, writing, or research agents.
• Host a streamable HTTP endpoint for a controlled team environment where each user supplies their own Kagi API key.

Safety and Privacy

Kagi MCP is a networked search and extraction server. Treat every query, URL, filter, lens ID, and extracted result as data sent to both the MCP client and Kagi's API. Avoid sending private URLs, credentials, internal tickets, customer names, or confidential research topics unless that use is approved for your Kagi account and model provider workflow.

When using HTTP mode, protect the endpoint like an API service. The server accepts bearer tokens and passes them to Kagi, so reverse proxies, platform logs, debugging middleware, and error reporting must not record authorization headers.

Duplicate Notes

Existing catalog entries cover no-key public search, scraping/crawling tools, and research agents. This entry covers the official kagisearch/kagimcp repository and kagimcp package, which specifically wrap the Kagi Search and Extract APIs for MCP clients.