google-ai-mode

This MCP server connects any compatible code agent (Claude, Cursor, Cline, Windsurf, Zed, etc.) to Google's AI Mode, enabling AI-synthesized research results with inline citations from 100+ sources. Once running, your agent can issue natural language research queries and receive a single, grounded answer that includes source citations and a clean Markdown format. The server handles the heavy lifting by launching a stealth browser, performing the Google AI Mode search, extracting the AI-generated answer and citations, and delivering the formatted result back to your agent. It supports follow-up context across queries and saves results for later reference, making it ideal for complex research tasks, technical comparisons, regulatory summaries, and product evaluations.

To use it, integrate the MCP server with your agent according to your workflow (e.g., Claude Code, Codex, Cursor, Windsurf, or Zed). The typical setup is to run the MCP via npx google-ai-mode-mcp@latest and point your agent to the provided mcpServers configuration. When a query is issued, the server automatically queries Google AI Mode, parses the synthesized answer, and returns a citation-rich, structured response suitable for immediate display or further processing by your agent.

Prerequisites:

• Node.js and npm installed on your system
• Access to run MCP servers and modify your agent's MCP configuration

Install and run using npx (no global install required):

1. Ensure you have an MCP config file in your agent environment, for example:

{ "mcpServers": { "google-ai-mode": { "command": "npx", "args": ["google-ai-mode-mcp@latest"] } } }

2. Run or reload your MCP configuration so the agent can discover the google-ai-mode server. If you prefer a direct one-off run, you can execute:

npx google-ai-mode-mcp@latest

3. (Optional) Update the package when needed:

npm update -g google-ai-mode-mcp

4. If your agent requires a specific environment (like headless browser settings), consult the README for CAPTCHA handling and browser profile guidance.

Tips and common considerations:

• CAPTCHA handling may occur on the first run as a browser profile is created. If you encounter issues, you can switch to a visible browser temporarily or use a non-headless mode to solve CAPTCHAs manually.
• Results are saved to platform-specific locations (Linux/macOS/Windows) as described in the Quick Start section of the README.
• The server enforces English results by default unless configured otherwise; clear the browser profile if language issues arise.
• If you encounter repeated CAPTCHAs, consider adding short delays between queries or using the explicit visible browser approach as documented.
• The MCP config shown above uses the canonical npx invocation; you can adapt the command/args to fit your agent’s preferred workflow or include additional runtime flags if necessary.