minimax_search

MiniMax Search MCP Server provides web search and batch browsing capabilities powered by the MiniMax LLM. It exposes two primary tools: search for parallel web search across multiple queries using Google via Serper, and browse for batch intelligent browsing of multiple URLs with content understanding from the MiniMax LLM. The server supports parallel execution for multiple queries, advanced Google search syntax, and auto fallback to other engines if a search fails. After starting the MCP server, clients can call search with a list of queries to obtain aggregated results, or call browse with a list of URLs and a question to obtain a comprehensive summary grounded in the fetched content.

Prerequisites:

• Access to an MCP-enabled environment and the uvx runner is available in your setup.
• Internet access to fetch the minimax_search repository (if installing from Git).

Install from Git (recommended):

uvx --from git+ssh://git@github.com:MiniMax-AI/minimax_search.git minimax-search

Install from a local path (development):

uvx --from /path/to/minimax_search minimax-search

Configuration (example client config):

{
  "mcpServers": {
    "minimax_search": {
      "command": "uvx",
      "args": [
        "--from",
        "git+ssh://git@github.com:MiniMax-AI/minimax_search.git",
        "minimax-search"
      ],
      "env": {
        "MINIMAX_API_KEY": "your_minimax_api_key",
        "SERPER_API_KEY": "your_serper_api_key",
        "JINA_API_KEY": "your_jina_api_key"
      }
    }
  }
}

Notes:

• Ensure you have the necessary API keys and they are kept secure.
• The MCP server exposes two tools: search and browse; call them via your MCP client as defined in the README.

Tips and common issues:

• Ensure API keys (MINIMAX_API_KEY, SERPER_API_KEY, JINA_API_KEY) are valid and not expired.
• If a search fails, the Auto Fallback feature will switch to other engines if configured, so provide multiple engines when available.
• For advanced search, leverage Google syntax like site:, intitle:, inurl:, and exact phrases to refine results.
• When using browse, the MiniMax LLM will read page content and generate a summarized answer; ensure the input URLs are accessible and not behind heavy authentication.
• If starting from a local path, ensure the path is absolute or correctly relative to your environment and that dependencies are installed.