# 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.