google-ai

This MCP server exposes Google AI capabilities (Imagen 3 image generation, Gemini language models, and related tools) to Claude Desktop via an MCP interface. Install the server as an MCP backend using npx and configure Claude Desktop to point at the google-ai MCP server. The available tools include imagen_generate_image to create images from prompts, gemini_generate_text to generate text using Gemini models, and placeholders for veo_generate_video and lyria_generate_music that are listed as coming soon. When using mock mode (USE_MOCK=true), the server will return sample responses without consuming Google Cloud quotas, which is useful for testing and integration checks. The configuration supports providing a service account JSON to authenticate with Vertex AI, along with project and location details.

Prerequisites:

• Node.js and npm installed on the machine running Claude Desktop
• Claude Desktop installed and configured to load MCP servers
• A Google Cloud project with Vertex AI APIs enabled and a service account key

Step-by-step installation:

1. Install and run the MCP package via npm/npx by adding the server configuration to your Claude Desktop config:

{
  "mcpServers": {
    "google-ai": {
      "command": "npx",
      "args": ["@stevekaplanai/google-ai-mcp@latest"],
      "env": {
        "USE_MOCK": "false",
        "GOOGLE_CLOUD_PROJECT": "your-project-id",
        "GOOGLE_CLOUD_LOCATION": "us-central1",
        "GOOGLE_APPLICATION_CREDENTIALS_JSON": "paste-json-here"
      }
    }
  }
}

2. Prepare Google Cloud credentials:

• Enable Vertex AI API in your project
• Create and download a service account key (as JSON) with the aiplatform.user role
• If using real credentials, paste the JSON into GOOGLE_APPLICATION_CREDENTIALS_JSON, escaping newlines as \n and placing all content on a single line

3. Update the Claude Desktop config file with the Google AI MCP server entry shown above (Windows or macOS paths provided in the README)
4. Restart Claude Desktop to apply the changes
5. Optionally enable mock mode by setting USE_MOCK to true for testing without API calls

Tips and tips:

• Keep service account keys secure; do not commit to version control.
• When pasting the credentials JSON, ensure newlines are escaped (\n) and the JSON is on a single line.
• If you encounter JSON parsing or authentication errors, verify the project ID, location, and that the Vertex AI API is enabled.
• The mock mode can help test prompts and tool invocation without consuming quota.
• The available tools support Imagen image generation (imagen_generate_image) and Gemini text generation (gemini_generate_text); later tools (veo_generate_video, lyria_generate_music) are labeled as coming soon.
• Monitor compatibility with Claude Desktop versions and update to the latest MCP package when issues arise.