MCP Gemini Server

What is MCP Gemini Server?

This server wraps the @google/genai SDK, exposing Google Gemini model capabilities as standard MCP tools. It allows other LLMs (like Cline) or MCP-compatible systems to leverage Gemini’s text generation, function calling, chat, file handling, and caching features.

How to use MCP Gemini Server?

Install via Smithery (npx -y @smithery/cli install @bsmi021/mcp-gemini-server) or manually (clone, npm install, npm run build). Configure your MCP client’s settings with the server’s path and required environment variables (GOOGLE_GEMINI_API_KEY, optionally GOOGLE_GEMINI_MODEL). Restart the client after configuration.

Key features of MCP Gemini Server

• Standard and streaming text generation (gemini_generateContent, gemini_generateContentStream)
• Function calling that lets Gemini request client-defined functions
• Stateful chat sessions with history management
• File upload, list, retrieve, and delete operations
• Cache content creation, listing, retrieval, update, and deletion

Use cases of MCP Gemini Server

• Integrate Gemini text generation into MCP-compatible assistants (e.g., Cline)
• Enable function calling workflows where Gemini invokes external tools
• Build conversational agents that maintain context across multiple turns
• Upload and manage files (images, documents) for multimodal prompts
• Optimize repeated prompt patterns with server-side content caching

FAQ from MCP Gemini Server

What runtime is required?

Node.js version 18 or later.

Where do I get an API key?

An API key from Google AI Studio (https://aistudio.google.com/app/apikey).

Can I use Vertex AI authentication?

No. The server only supports Google AI Studio API keys; file handling and caching features are not compatible with Vertex AI credentials.

How do I set a default Gemini model?

Set the GOOGLE_GEMINI_MODEL environment variable (e.g., gemini-1.5-flash). If not set, the modelName parameter becomes required on applicable tool calls.

What error types does the server return?

Errors are returned as McpError objects with an error code (InvalidParams, InternalError, etc.), a message, and optional details for troubleshooting (e.g., safety block reasons).