Search is implemented; the three other Kagi capabilities are listed as roadmap items with a note that they cannot be used yet, which is more candour than most repositories offer. Treat this as a way to route search through Kagi rather than as coverage of the Kagi product.
A TypeScript MCP server that puts Kagi's search behind a single tool, so an assistant can search the web through a Kagi account rather than through an ad-funded engine.
- A web search run against Kagi's API, taking a query string and an optional limit, returning Kagi's results — `kagi_search`
- Summarisation, FastGPT answers and enriched news are named in the roadmap as `kagi_summarize`, `kagi_fastgpt` and `kagi_enrich`; the README states they are not implemented and cannot be called
A Kagi API key in `KAGI_API_KEY`, either in a `.env` file at the repository root or in the server's `env` block. The Kagi API is described in the README as being in private beta, so access to it is the gate rather than the server. The npm package name is `kagi-server` (0.1.0) and it is marked private; install from a clone with `npm install` and `npm run build`, or let the Smithery installer write the Claude Desktop entry.
One command plus a key — npx @smithery/cli install kagi-server --client claude, then supply credentials