前几天我分享了一个用于高效抓取网页的 Fetcher MCP: https://v2ex.com/t/1119976
今天分享 Fetcher MCP 的好搭档 G-Search MCP: G-Search
G-Search MCP 一款强大的谷歌搜索 MCP 服务器，可实现同时使用多个关键词进行并行搜索。
用法
npx -y g-search-mcp
在第一次使用时，如果遇到问题，可能需要先执行一下以下命令:
npx playwright install chromium
优势
- 并行搜索：支持在 Google 上同时使用多个关键词进行搜索，提高搜索效率。
- 浏览器优化：在单个浏览器实例中打开多个标签页进行高效的并行搜索。
- 自动验证处理：智能检测 CAPTCHA 验证码，并在需要用户验证时启用可见浏览器模式。
- 模拟用户行为：模拟真实用户浏览模式，降低被搜索引擎检测的风险。
- 结构化数据：以 JSON 格式返回结构化的搜索结果，方便处理和分析。
- 可配置参数：支持各种参数配置，如搜索结果数量限制、超时设置、区域设置等。
用法展示
1. deep search 模拟
system prompt:
Please conduct in-depth research on the topic provided by the user, using the MCP tools g-search:search and fetcher:fetch_urls, and produce a detailed research report.
Research Steps:
Topic Understanding and Keyword Generation: Thoroughly understand the research topic provided by the user. Analyze the core concepts and related aspects of the topic from different perspectives. Based on this analysis, generate a comprehensive set of search keywords in both Chinese and English, aiming to cover all dimensions of the topic.
Multi-Round Google Search:
First Round Search: Use the keywords generated in step 1 to call the g-search:search tool for parallel multi-keyword searches. If necessary, perform searches using both Chinese and English keywords separately to obtain more comprehensive information.
Result Filtering and Link Extraction: Carefully review the Google search results from each round and filter out webpage links that are highly relevant to the research topic. Focus on sources that are authoritative and information-rich.
Webpage Fetching: Use the fetcher:fetch_urls tool to batch-fetch the content of the webpage links filtered in step 2.2.
Content Analysis and Information Integration: Conduct in-depth analysis of the fetched webpage content. Extract key information, core viewpoints, and important data. Integrate information from different webpages, de-duplicate, compare, and verify it to form a systematic understanding of the research topic.
Information Evaluation and Iteration Decision: Evaluate whether the currently collected and analyzed information is sufficient to support a comprehensive research report and draw reliable conclusions.
If Information is Insufficient: Analyze the information gaps and identify directions for further research. Based on the existing information and analysis, optimize or expand the search keywords, or adjust the search strategy (e.g., focus more on specific types of sources, delve deeper into specific aspects of information). Then return to step 2 to conduct the next round of searches and information collection.
If Information is Sufficient: Proceed to step 5.
Research Report Generation: Based on all the collected and analyzed information, write a well-structured and detailed research report. The report should include:
Introduction: Clearly state the research topic and purpose.
Research Methodology: Briefly describe the search and information analysis methods used.
Research Findings: Present the key information, core viewpoints, data, and facts discovered during the research process in detail. Organize them by topic and perspective if possible.
Conclusions and Recommendations: Based on the research findings, summarize the main conclusions and propose corresponding recommendations or future outlooks based on the research results.
References: List all cited webpage links to ensure the traceability of information sources.
Tool Usage Requirements:
Be sure to use the g-search:search tool for Google searches and the fetcher:fetch_urls tool for fetching webpage content.
When calling the tools, please ensure that all necessary parameters are provided and set parameter values reasonably according to the actual situation (e.g., search keywords, fetching links, etc.).
In the information evaluation and iteration decision stage, demonstrate intelligent analysis and judgment capabilities, and effectively plan the subsequent research directions.
The final research report should be complete in structure, accurate in content, and fluent in language.
user prompt:
调研目前编程能力最强的非思考类型的大语言模型前三名分别是什么，并给出事实依据
输出效果：
过程说明：
-
AI 首先根据调研主题生成了多角度多语言的搜索关键词：
- best programming LLMs
- top code generation models
- 非思考型大语言模型 编程能力排行
- 编程能力最强 LLM
- ...
- 然后调用 G-Search MCP 并发同时进行搜索
- 拿到搜索结果后，AI 会挑选最相关的网页 url 集合，然后调用 fetcher 批量抓取内容详情
- AI 对信息进行整合进行评估，发现资料不足以得出结论
- 然后重复上述过程
- 最终得到足够多的资料后输出结论报告
目前尚无回复
Select Language