kagi-ken-mcp
@czottmann
Unofficial MCP server for working with Kagi *without* API access (you'll need to be a customer). Searches and summarizes. Uses Kagi session token for easy authentication. Works well with your agent.
概要
What is kagi-ken-mcp?
kagi-ken-mcp is a lightweight Node MCP server that wraps the kagi-ken package to provide access to Kagi.com services—Search and Summarizer—using a Kagi session token instead of an API key. It is built for users who have a Kagi subscription but lack invite-only API access.
How to use kagi-ken-mcp?
Requires Node.js 22+. Obtain your Kagi session token from your Kagi account settings (Session Link) and save it to ~/.kagi_session_token or set as the KAGI_SESSION_TOKEN environment variable. Add the server to Claude Desktop or Claude Code via the provided JSON configuration, then invoke its two tools (kagi_search_fetch and kagi_summarizer) for searches or URL summarization.
Key features of kagi-ken-mcp
- Search: fetch web results via Kagi Search with concurrent query support.
- Summarize: generate summaries from URLs, with customizable output type and language.
- Session token provided via environment variable or a local file.
- Comprehensive error handling: timeouts, input validation, and graceful formatting.
- No separate API key required — works with an existing Kagi session.
Use cases of kagi-ken-mcp
- Perform web searches from within LLM agents like Claude without needing Kagi API access.
- Summarize webpages, videos, or audio content using Kagi’s Summarizer.
- Integrate Kagi’s search and summarization into custom workflows or automation tools.
FAQ from kagi-ken-mcp
How does kagi-ken-mcp differ from the official Kagi API?
The official Kagi API requires an API key (currently invite-only). kagi-ken-mcp uses your existing Kagi session token, making it accessible to any Kagi subscriber without separate API approval.
What runtime does kagi-ken-mcp require?
Node.js version 22 or later is required.
Where does my session token come from and how is it stored?
You copy it from the Session Link in your Kagi account settings. It can be stored in the KAGI_SESSION_TOKEN environment variable or in a file at ~/.kagi_session_token. The server checks the env var first, then falls back to the file.
What transport and authentication does kagi-ken-mcp use?
It uses the standard MCP transport (stdio) and authenticates via the session token. No extra auth mechanism is needed.
Are there any known limitations?
Each search query has a 10-second connection timeout. The server only supports the two tools listed (search and summarizer) and does not expose other Kagi API endpoints.
