Kagi MCP Server
Search Kagi and extract full web page content as markdown for your AI models.
Data last scanned today · Reviewed yesterday
Overview
This MCP server provides a direct interface to the Kagi search API, allowing AI assistants to perform web, news, video, and podcast searches. It includes specific tools for web information retrieval, such as Kagi lens support and filtered search results. Beyond basic searching, the server includes a tool for fetching complete page content and converting it into markdown format. Users can either run the server locally using the uv package manager or connect to the official hosted endpoint provided by the project authors.
Our verdict
With a security score of 30/100, this tool carries a high risk profile. The lack of native authentication methods means that anyone with access to your configuration can utilize your Kagi API key, and the absence of a read-only mode requires careful handling of permissions. While useful for researchers requiring Kagi search data, it should be deployed with caution until proper authentication protocols like OAuth are implemented.
- Supports multiple search categories including podcasts and images
- Provides direct markdown extraction for web page content
- Zero direct dependencies reduce the supply-chain attack surface
- Active development with commits within the last ten days
- No authentication method forces reliance on local configuration files
- No read-only mode restricts control over server capabilities
- Project is community-maintained rather than vendor-managed
Tools
| Tool | Description | Risk |
|---|---|---|
| kagi_search_fetch | web, news, videos, podcasts, and image search with optional page extracts, filters, and Kagi lenses. | low |
| kagi_extract | fetch a page's full content as markdown. | low |
Compatibility
| Client | Local | Docker | Remote | Read-only |
|---|---|---|---|---|
| Claude Desktop | ||||
| Cursor | ||||
| VS Code | ||||
| Windsurf | ||||
| ChatGPT |
Frequently asked questions
›Do I need to install anything if I want to use the hosted version of the server?
No, you do not need to install anything for the hosted server; you can simply point your HTTP-capable MCP client to https://mcp.kagi.com/mcp and authenticate using your Kagi API key via Bearer token.
›What prerequisites are needed for a local installation of the server?
You must have a valid Kagi API key and the uv package manager installed on your system to use the recommended uvx install path.
›Can I use OAuth2 to authenticate with the server?
No, OAuth2 is not currently supported; you must use an API key obtained from your Kagi dashboard passed via Bearer HTTP authentication.
›Are the kagi_fastgpt and kagi_summarizer tools currently available?
No, those specific tools were removed in a recent update, though their return is planned for a future release.
›How can I configure the server to hide specific search parameters from the LLM?
You can use the KAGI_HIDDEN_PARAMS environment variable, set to a comma-separated list of the search parameters you wish to exclude from the LLM-facing schema.
›Does the self-hosted HTTP mode support multiple users?
Yes, the HTTP mode is multi-tenant, meaning a single instance can serve multiple users because each request provides its own API key via the Authorization header.
Alternatives
Badge
Maintain this server? Add the live badge to your README.