Gemini

You can run the Gemini MCP Server to access Google's Gemini API through a Model Context Protocol interface. It supports local STDIO workflows for direct integration and HTTP transport for remote access, enabling you to perform web searches, content generation, and Gemini model tasks from your LLM or tooling setup.

How to use

Choose your transport mode and start the server, then connect to it from your MCP client. In STDIO mode, you run the server locally and communicate via standard input/output. In HTTP mode, you run the server in streamable-http mode and connect over HTTP with an Authorization header.

STDIO mode lets you connect directly within your application or tooling that supports MCP STDIO. Start the server with your API key and the stdio transport. This mode uses the GEMINI_API_KEY environment variable for authentication.

How to install

Prerequisites: you need Python and the pip package manager available on your system.

Install the Gemini MCP Server package from its source control using pip.

pip install git+https://github.com/philschmid/gemini-mcp-server.git

Additional notes

Deployment to remote environments is supported. You can run the server in Google Cloud Run to make it accessible to any client. When running remotely, you typically deploy the service with unauthenticated access but protect it with an Authorization header at the client side.

Security: STDIO mode relies on the GEMINI_API_KEY environment variable. HTTP mode relies on a Bearer token provided via the Authorization header.

Examples show how to configure your MCP client to connect in both modes. Use the appropriate URL or command as shown below.

Security considerations for remote access

If you expose the HTTP endpoint, ensure requests include a valid Authorization header. You may choose to deploy behind a gateway or require authentication to limit access to trusted clients.

Tools and capabilities

web_search — Performs a web search using Gemini and returns synthesized results with citations. Use this to retrieve up-to-date information and attach sources when presenting results.

use_gemini — Delegates a task to a Gemini 2.5 model (Pro or Flash). Provide a prompt and optionally select the Gemini model to use.

Available tools

web_search

Performs a web search using Gemini and returns synthesized results with citations.

use_gemini

Delegates a task to a Gemini 2.5 model (Pro or Flash).