gemini-cli/docs/tools/web-fetch.md

1.5 KiB

Web Fetch Tool

This document describes the web_fetch tool.

web_fetch

  • Purpose: Processes content from one or more URLs (up to 20) embedded in a prompt. This tool is ideal for tasks that require summarizing, comparing, or extracting specific information from web pages.
  • Arguments:
    • prompt (string, required): A comprehensive prompt that includes the URL(s) (up to 20) to fetch and specific instructions on how to process their content. For example: "Summarize https://example.com/article and extract key points from https://another.com/data". The prompt must contain at least one URL starting with http:// or https://.
  • Behavior:
    • The tool sends the prompt and the specified URLs to the Gemini API.
    • The API fetches the content of the URLs, processes it according to the instructions in the prompt, and returns a consolidated response.
    • The tool formats the response, including source attribution with citations, and returns it to the user.
  • Examples:
    • Summarizing a single article:
      web_fetch(prompt="Can you summarize the main points of https://example.com/news/latest")
      
    • Comparing two articles:
      web_fetch(prompt="What are the differences in the conclusions of these two papers: https://arxiv.org/abs/2401.0001 and https://arxiv.org/abs/2401.0002?")
      
  • Important Notes:
    • This tool relies on the Gemini API's ability to access and process the given URLs.
    • The quality of the output depends on the clarity of the instructions in the prompt.