SerpKite

What SerpKite MCP server does

SerpKite MCP server gives MCP-compatible agents access to Google-backed discovery and public web retrieval. Search results are returned as Markdown intended for language-model use. The service can answer general web queries, find news articles, locate places and ratings, search academic papers, and query patents, products, images, and videos.

It also supports content-oriented workflows. An agent can convert a public HTML page or PDF to Markdown, extract content from multiple URLs, discover URLs on a site, and start or inspect a crawl of public pages. Only public, logged-out pages are fetched.

How it works

The repository documents a hosted remote service rather than implementing the API locally. Clients connect to https://api.serpkite.com/v1/mcp using stateless Streamable HTTP. The endpoint supports MCP protocol versions 2025-06-18, 2025-03-26, and 2024-11-05.

Authentication can use MCP OAuth 2.1 with the mcp scope, or an API key sent as a bearer token. OAuth-capable clients open a SerpKite sign-in and approval flow; the approved connection is represented by an API key on the account. For clients without MCP OAuth, configure a key from the SerpKite account service. Credits do not expire, and failed or empty searches are free, while tool pricing follows the corresponding REST endpoints.

Tools and capabilities

The documented tool set includes:

• search for Google web results, direct answers, knowledge-graph data, and related questions
• news, maps, and scholar for news, places, and academic research
• patents, shopping, images, and videos for specialized searches
• autocomplete for query suggestions
• webpage for converting public HTML or PDF content to Markdown
• extract for reading multiple URLs and optionally selecting passages related to a query
• map, crawl, and crawl_result for site URL discovery and public-page crawling

SerpKite MCP server is suited to research agents, web-answer workflows, source collection, and tasks that need page content rather than only search-result links. The available tools can also support location research, literature discovery, product comparisons, and public website inspection.

Setup and configuration

Use the remote URL directly in clients that support Streamable HTTP. Claude Code can add it with claude mcp add --transport http serpkite https://api.serpkite.com/v1/mcp, followed by OAuth authentication or an authorization header. Cursor uses an MCP JSON entry with the endpoint and SERPKITE_API_KEY; the repository also includes a Cursor plugin that reads that variable without storing the key. VS Code accepts an HTTP server entry with a password-protected input, and Cline has a separate remote-connection configuration.

Claude Desktop and other stdio-oriented clients can use the third-party mcp-remote Node.js bridge. ChatGPT requires developer mode, a connector using the endpoint, and OAuth authentication. Keep API keys out of source control. The repository includes curl requests for sending initialize and tools/list to verify connectivity without running a search.

Limitations and notes

The API implementation is proprietary, while the repository documentation and configuration are MIT licensed. The service fetches only public, logged-out pages, so it is not intended for authenticated or private web content. Google trademarks belong to Google LLC, and the service states that it is not affiliated with Google.