Searches the web and reads links

This is a Model Context Protocol (MCP) server project.

• Calls the Bing Search API with the provided query.
• Processes each URL from the search results:
  • Gets the content of the page
  • Converts it to Markdown using Markitdown
  • Parses out links separately. Caches a unique hash to associate with each link.
  • (Optional, on by default) Uses sampling to select the most important links to return.
  • (Optional, on by default) Filters out the Markdown content to the most important parts.
• Returns the processed content and links as a LLM-friendly string.

• Takes a list of hashes (which originate from the search tool).
• For each hash gets the corresponding URL from the local cache.
• Then does the same processing as search for each URL and returns a similar LLM-friendly string.

Simply run:

make

To create the virtual environment and install dependencies.

Create a .env file based on .env.sample and populate it with:

• BING_SEARCH_API_KEY
• ASSISTANT__AZURE_OPENAI_ENDPOINT - This is necessary if you want to post process web content.

Use the VSCode launch configuration, or run manually:

Defaults to stdio transport:

uv run mcp-server-bing-search

For SSE transport:

uv run mcp-server-bing-search --transport sse --port 6030

The SSE URL is:

http://127.0.0.1:6030/sse

To use this MCP server in your setup, consider the following configuration:

{
  "mcpServers": {
    "mcp-server-bing-search": {
      "command": "uv",
      "args": ["run", "-m", "mcp_server_bing_search.start"]
    }
  }
}

{
  "mcpServers": {
    "mcp-server-bing-search": {
      "command": "http://127.0.0.1:6030/sse",
      "args": []
    }
  }
}