What is Gsearch?
Gsearch is a self-hosted MCP server that lets AI agents search Google and receive results as typed JSON. It is designed for people building or using agents in MCP-compatible clients, including Claude Desktop and Cursor. The server uses a real Firefox browser and can run on your own IP or through a proxy. Setup requires Python 3.12 or later and a one-time browser download.
Gsearch Features:
Google Results: The google_search tool returns Google results with titles, snippets, domains, dates, display URLs and pagination details.
Search Page Data: When present on the results page, the response includes AI Overview text and sources, People also ask questions, related searches, ads and sitelinks.
Page Reading: Agents can use fetch_page to read a result or read_top to retrieve the text of top pages alongside search results.
Link Resolution: The resolve_links tool converts Google's result redirects into destination URLs, while fetch_page can follow an individual result link.
Browser Search: Gsearch runs Google's page in Camoufox, a Firefox browser, and can use the local IP or a configured proxy.
Structured Errors: Blocks and other failures return named errors, such as captcha or rate_limited, with a retry time when available instead of empty results.
Gsearch Benefits:
Direct Google Coverage: Agents can work with results from Google's index rather than relying on a different search provider's index.
Fewer Manual Steps: Search results and page text can be collected through MCP calls, reducing the need to copy links and content into an agent by hand.
Clearer Failures: Structured error responses help an agent distinguish a search block from a genuinely empty results page.
Flexible Hosting: Running the browser on your own machine or server gives you control over where searches are made.
No Search API Key: You can try Google searches without obtaining an API key or paying a per-query search provider.
Use Cases:
Agent Research: Ask an MCP-connected agent to find current information on a topic and return a short list of Google results.
Question Discovery: Gather People also ask questions and related searches to identify follow-up queries for a research task.
Source Summaries: Have an agent search for documentation, read a result page and summarize its contents.
Domain Research: Search for pages from a chosen site, such as finding recent documentation or articles on a specific domain.
Search Result Review: Inspect result details, AI Overview sources and destination URLs when checking what Google surfaces for a query.

