Serpapi MCP

What serpapi/serpapi-mcp MCP server does

The serpapi/serpapi-mcp MCP server exposes SerpApi search and data extraction through the Model Context Protocol. Its central search tool supports multiple search engines, including Google, Bing, Yahoo, DuckDuckGo, YouTube, eBay, and other engines listed by SerpApi. Search results may include organic listings, answer boxes, news, images, and shopping data, with result formatting detected automatically.

The server can also retrieve information through ordinary search queries, such as location-based weather forecasts or company stock data. It supports JSON responses by default, with complete and compact modes, and can return Markdown when params.output is set to md. Optional search_table and search_dashboard tools provide interactive result displays in MCP hosts that support MCP Apps.

This places the serpapi/serpapi-mcp MCP server in the 🔎 Search & Data Extraction category. It fits agents that need current search results or engine-specific data without implementing separate clients for each supported search provider.

How it works

Clients call search with a params object. The required params.q field is used for standard queries, while some engines use fields such as search_query, term, or _nkw. The params.engine value selects the SerpApi engine and defaults to google_light when omitted. Other parameters can specify location, domain, and engine-specific search options.

Engine parameter schemas are available through MCP resources at serpapi://engines and serpapi://engines/<engine>. Clients with argument-completion support can request engine-name suggestions for the resource URI. The response includes the result in MCP structured content and text; JSON results contain serialized JSON, while Markdown results contain formatted Markdown.

Setup and configuration

The hosted service is available at https://mcp.serpapi.com/mcp. Clients can authenticate with an Authorization: Bearer YOUR_API_KEY header, which keeps the key out of the URL, or use the path form /YOUR_API_KEY/mcp for clients that cannot set headers. A SerpApi API key is needed for search and the App tools, but not for connecting, listing tools, or reading resources.

For local deployment, clone the repository, run uv sync, and start it with uv run src/server.py. A local HTTP endpoint can then be configured in an MCP client. The repository also provides a src/stdio.py entry point for hosts that launch MCP servers as subprocesses. The project requires Python 3.13 or newer. Claude Desktop users can install the released .mcpb bundle, which asks for the API key and runs the server locally over stdio.

Tools and capabilities

The serpapi/serpapi-mcp MCP server provides:

• A general search tool covering all SerpApi engines and engine-specific parameters.
• Engine schema resources for discovering supported parameters.
• Complete or compact JSON output modes.
• Markdown output for lower-token result transfer.
• Optional search_table and search_dashboard interactive UI tools.
• Search-based access to weather, stock, news, image, shopping, and other supported result types.

Limitations and notes

Results depend on the SerpApi API and the selected engine’s available parameters. The server does not expose one fixed schema for every search: parameters vary by engine, so clients should consult the engine resources or SerpApi’s search API reference. An API key is required for actual searches and App tools. Markdown output is unchanged by the JSON mode setting, and compact mode removes JSON metadata rather than changing the underlying search request.