Overview
This MCP server implements a local MCP endpoint for the RAG Web Browser Actor, designed to give fast responses to AI agents and LLMs by enabling real-time web browsing capabilities within a RAG workflow. It supports web search by querying Google Search or fetching direct URLs, scraping and returning cleaned page content in Markdown, and can also fetch a single URL directly. It runs locally and communicates with the RAG Web Browser Actor in Standby mode, exchanging search queries and content extractions. The server exposes a single tool named search with configurable arguments: query (required string), maxResults (optional number, default 1), scrapingTool (optional string, default raw-http; values browser-playwright or raw-http), outputFormats (optional array of strings, default [markdown]), and requestTimeoutSecs (optional number, default 40, min 1, max 300). There is also standard input/output (stdio) MCP integration for seamless communication with AI clients. The project notes that this local MCP server is deprecated in favor of mcp.apify.com, with guidance on migration and links to hosted endpoints and client tools.
Features
Tools: search
Query Google Search OR fetch a direct URL and return cleaned page contents.
Output formats
Supports outputFormats (text, markdown, html) with default markdown.
Scraping tool selector
scrapingTool option: raw-http or browser-playwright to handle static vs JS-heavy sites.
Configurable max results
maxResults controls number of results (default 1, ignored for direct URL queries).
Request timeout
requestTimeoutSecs sets total server-side and client wait budget (default 40s, 1-300).
Local MCP stdio
Local MCP integration via standard input/output for AI clients.
Standby mode communication
Operates with the RAG Web Browser Actor in Standby mode to exchange queries and content.
Who Is This For?
- AI agents:Provide fast web browsing results and content extraction for RAG workflows.
- LLMs:Enable LLMs to perform web searches and fetch content in Markdown for analysis.
