mcp-webresearch

The MCP Web Research Server provides real-time web research capabilities for Claude via three main tools: search_google to perform Google searches and extract results, visit_page to navigate to a URL and extract page content (with an optional screenshot), and take_screenshot to capture the current page view. A prebuilt agentic-research prompt is available to guide Claude through systematic, source-cited internet research; you can access it in Claude Desktop by selecting the webresearch integration under the Paperclip menu. The server also maintains a Research Session resource that records queries, visited pages, extracted content, and timestamps, enabling you to review the research history and results later.

To use it, configure Claude Desktop to start the MCP server automatically, then start a chat and prompt Claude to perform web research. When you call search_google, pass a query to retrieve and rank sources; with visit_page, provide a URL and an optional flag to capture a screenshot; and with take_screenshot, request a snapshot of the current page when needed. For deeper workflows, use the agentic-research prompt to structure the research process, ensuring high-quality sources are cited with URLs throughout the session.

Prerequisites:

• Node.js (>= 18) with npm and npx available
• Claude Desktop app installed

Step-by-step installation:

1. Install Node.js from https://nodejs.org/ (ensure npm and npx are available on the command line).
2. In Claude Desktop, download and set up the MCP server by adding the following configuration to claude_desktop_config.json (Mac path example provided):

{
  "mcpServers": {
    "webresearch": {
      "command": "npx",
      "args": ["-y", "@mzxrai/mcp-webresearch@latest"]
    }
  }
}

3. Restart Claude Desktop or refresh MCP connections so the web research server is automatically started when needed.
4. Open a chat with Claude and initiate a web research task; Claude will use the MCP server to perform searches, visit pages, and capture screenshots as requested.

Notes and tips:

• This server is used via npx, so an internet connection is required to fetch the latest package on first run.
• The server exposes two MCP resources: captured webpage screenshots and the research session history. Access them through Claude Desktop as part of the Paperclip integration.
• If you encounter issues, check Claude Desktop MCP logs at: tail -n 20 -f ~/Library/Logs/Claude/mcp*.log (macOS). Ensure Playwright is installed as a dependency; it will be auto-managed by the package.
• You can customize prompts with agentic-research to steer the research direction and ensure citations accompany findings.
• The workflow is pre-alpha; expect occasional bugs and verify sources when performing critical research.