Kagi Search
FreeKagi 官方集成，使用 Kagi 的搜索 API 进行网页搜索。
FreeFree tier
About Kagi Search
The Official Model Context Protocol (MCP) server for Kagi Search. It exposes two main tools: kagi_search_fetch for searching the web, news, videos, podcasts, and images (with optional page extracts, filters, and Kagi lenses), and kagi_extract for fetching a page's full content as markdown. The server can be used via a hosted endpoint at https://mcp.kagi.com/mcp (no installation required) or run locally with uvx. Requires a Kagi API key for authentication. Supports configuration of timeouts, retries, and search parameters via environment variables. Compatible with various MCP clients including Claude Desktop, Claude Code, Codex CLI, Smithery, Kiro, and OpenCode.
Key Features
Provides two main tools: kagi_search_fetch (web, news, videos, podcasts, images search with optional page extracts and Kagi lenses) and kagi_extract (full page content as markdown)
Hosted server available at https://mcp.kagi.com/mcp – no local installation required
Local installation via uvx with simple CLI setup for Claude Desktop, Claude Code, Codex CLI, Smithery, Kiro, and OpenCode
Configurable environment variables for timeouts (search default 10s, extract default 30s), retries (default 2), and hidden search parameters
Supports Bearer HTTP authentication with a Kagi API key (OAuth2 on roadmap)
Open-source repository on GitHub under kagisearch/kagimcp
Pros & Cons
Pros
- Official and up-to-date integration with Kagi Search, maintained by the Kagi team
- Flexible deployment: hosted server for zero-install or self-hosted for full control
- Supports a wide range of search types (web, news, video, podcast, image) with filters and lenses
- Extracts page content as clean markdown, useful for feeding into LLMs
- Well-documented configuration options for timeouts, retries, and search parameters
- Compatible with multiple popular MCP clients out of the box
Cons
- Requires a Kagi API key, which is part of Kagi's paid search subscription (no free tier beyond limited searches)
- OAuth2 authentication is not yet supported – only API key bearer authentication
- Previous fastGPT and summarizer tools have been removed and are only planned for a future release
- Limited to Kagi's search index and ecosystem
Best For
Integrate Kagi's search capabilities into AI assistants and LLM workflows via the Model Context ProtocolRetrieve and extract full web page content as markdown for analysis or processingPerform multi-media searches (news, videos, podcasts, images) from AI agentsEnable real-time web search and content extraction for MCP-compatible clients
FAQ
What tools does the Kagi MCP server provide?
It provides two tools: kagi_search_fetch for web, news, videos, podcasts, and image search with optional page extracts, filters, and Kagi lenses; and kagi_extract for fetching a page's full content as markdown.
Do I need to install anything to use it?
No. You can use the hosted server at https://mcp.kagi.com/mcp with your Kagi API key via Bearer HTTP authentication. Alternatively, you can run it locally with uvx for a self-hosted setup.
How do I authenticate?
Use your Kagi API key from the dashboard. Pass it via the Authorization header as 'Bearer <API_KEY>'. OAuth2 is not yet supported but is on the roadmap.
What happened to the fastGPT and summarizer tools?
They were removed from the current release but are planned to return in a future release.
Which MCP clients are supported?
The server works with Claude Desktop, Claude Code, Codex CLI, Smithery, Kiro, and OpenCode, among other HTTP-capable MCP clients.